Spectrum of Engineering Sciences

GEOTECHNICAL SITE INVESTIGATION AND SEDIMENTATION ANALYSIS FOR THE OPTIMIZATION OF SAND TRAP DESIGN AT THE KOTO HYDRO POWER PROJECT, DIR LOWER, KHYBER PAKHTUNKHWA, PAKISTAN

Jehanzeb Khan1 ⃰ — 2025-08-20

The Koto Hydropower Project in Koto Town, District Lower Dir, Pakistan, focuses on harnessing Panjkora River energy, with this study detailing the geotechnical investigation informing its sand trap design. The site lies within the geologically complex Kohistan Complex, featuring major thrust faults, diverse crystalline rocks (granites, gneisses, schists), and varied weathering/metamorphism. Rock properties, including variable permeability due to foliated schists and weathering, plus fault-induced fractured zones, critically impact strength, permeability, and localized instability. A geotechnical investigation primarily utilizes data from four boreholes directly relevant to the sand trap investigation and in-situ SPT/CPT, revealed permeability from 10−3 to 10−7 cm/s and shear strength from 20° to 40° (average 30°), with lower strengths in weathered/fractured areas. These findings were crucial for optimizing the sand trap's design, influencing its sizing, hydraulic design for flow variations, foundation stability near faults, reinforcement in low-strength areas, and material selection for durability against erosion/clogging. The optimized sand trap ensures efficient sediment removal, crucial for the hydropower plant's longevity, operational efficiency, and reduced maintenance costs.

Keywords -(Dir Lower, Koto Hydropower Project, Sand Trap)

BIOCHAR-BASED CONCRETE BLOCKS FROM RICE HUSK: A SUSTAINABLE SOLUTION FOR LOW-CARBON CONSTRUCTION

admin admin — 2025-08-20

This study investigates rice husk biochar (RHB) as a partial cement replacement in concrete blocks to enhance sustainability and reduce the environmental footprint of construction materials. RHB, produced via pyrolysis of agricultural residues, offers high porosity, silica content, and carbon sequestration potential, making it a promising additive. Experimental mixes were prepared with 5–30 wt. % RHB, using Ordinary Portland Cement, natural siliceous sand, and tap water. Tests included compressive and flexural strength, ultrasonic pulse velocity (UPV), water absorption, permeability, and thermal conductivity. Results showed that replacing 20–30 wt. % cement with RHB increased compressive strength by up to 15% when pre-mixed in a water-super plasticizer solution, while 20% replacement improved flexural strength by ~30%. UPV readings (~3 km/s) indicated no significant compromise in material integrity. Pre-soaked biochar reduced water absorption by up to 41% at 30% replacement, due to reduced porosity. Thermal conductivity decreased steadily with higher biochar content, confirming its insulation benefits. However, replacement levels above 15% in load-bearing applications led to strength reductions, indicating the need for optimal proportioning. Environmental assessment confirmed substantial carbon sequestration potential, with literature reporting up to 870 kg CO₂-eq mitigation per ton of dry feedstock. Additionally, using RHB supports waste valorization, reduces landfill pressure, and lowers embodied carbon. The mix designs required no specialized production methods, enabling immediate adoption in industry. These findings demonstrate that RHB-based concrete blocks offer mechanical adequacy, improved thermal performance, and significant environmental benefits, making them suitable for non-load-bearing and energy-efficient applications. Future work should optimize mixes for structural uses and assess long-term performance under real-world conditions.

Keywords Rice Husk Biochar, Sustainable Concrete, Carbon Sequestration, Thermal Insulation, Eco-Friendly Construction, Green Building Materials

A REVIEW OF MACHINE LEARNING ALGORITHMS TO MINERAL EXPLORATION AND MAPPING

Shakir Ullah — 2025-08-17

The current breakthroughs in smart mining offerings have brought a new wave of real-time data production and analyses, and the mining sector has leaped forward embracing machine learning (ML) to streamline their activities, enhance safety, and increase sustainability. This review examines 87 new publications, and a careful study of 42 significant papers to examine how ML is being used in several mining disciplines, including mineral exploration, ore grade modeling, process optimization, and environment management. The results point out the fact that the ML research is highly focused on the application of surface mining where numerous challenges and opportunities are built on complexity and abundance of data. Such techniques as deep neural networks (DNNs) and support vector machines (SVMs) are popular because they show good results in predictive maintenance, ore classification, and yield optimization, but techniques such as ensemble methods and reinforcement learning are becoming increasingly popular because they are more adaptable. Though the classic criteria of evaluation such as the robustness of regression tend to be widespread, more sophisticated tools such as the cross-validation and confusion matrices are on the rise. Data heterogeneity, model transparency, and data incorporation of real-time sensor data continue to remain a problem. The upcoming studies are recommended to focus on hybrid solutions that combine ML with physics-based models, exploit edge computing to get on-the-fly realizations, and resolve the ethical aspect of AI automation. All in all, the review highlights the revolutionary properties of ML in the mining field and the necessity of a more coordinated work of data scientists, engineers, and stakeholders to facilitate the development of efficient, smart, and sustainable mining trends.

Keywords -(Smart mining, machine learning, deep neural networks, support vector machines, predictive modeling, mineral extraction, real-time data analytics).

https://doi.org/10.5281/zenodo.16890844

EFFICIENT IMAGE DESCRIPTOR GENERATION USING CNN ARCHITECTURES FOR ENHANCED IMAGE RETRIEVAL

Muhammad Huzaifa Rashid — 2025-08-04

Machine learning algorithms are widely employed in image classification tasks to extract and represent discriminative features from images. In this study, we present an efficient approach for generating image descriptors using Convolutional Neural Network (CNN) architectures, including GoogleNet, Inception V3, and DenseNet-201. These networks are leveraged to capture both texture and object-level features, which are further encoded through three color channels to enhance image retrieval performance while maintaining an optimal response time. When images are processed through the hierarchical layers of the CNNs, distinctive feature representations (signatures) are produced. These signatures are subsequently used to construct a new matrix that effectively encodes spatial relationships, color attributes, and latent patterns, thereby providing a more comprehensive representation of image content. The proposed CNN-based method was evaluated on four benchmark datasets: Corel-1K, CIFAR-10, 17-Flowers, and ZuBuD. Among the tested architectures, DenseNet-201 achieved the best performance on the CIFAR-10 dataset, which contains images of diverse categories and varying sizes, demonstrating superior accuracy compared to GoogleNet and Inception V3.

IDENTIFICATION AND CLASSIFICATION OF FOODBORNE DISEASE OUTBREAKS

Ali Zain — 2025-08-02

Foodborne disease is commonly caused by consuming contaminated food and beverages so the identification and classification of foodborne disease outbreaks is necessary to prevent and reduce the risk of illness and death. The purpose of this research is to identify the causative agents of disease as soon as possible to improve the food safety to prevent from illnesses and deaths. The useful patterns have been identified with analysis on dataset and also determine the large number of outbreaks occurs in year, food, location and species. The classification is done in Decision Tree, Naïve Bayes and Random Forest classifiers. The experiments on the dataset have proven the efficiency of purposed approach for identification and classification of outbreak patterns.

IMPACT OF NANO-SILICA ON THE MECHANICAL BEHAVIOR OF GEOPOLYMER CONCRETE

Muhammad Rashid Naveed — 2025-08-02

There is an enormous use of Portland cement concrete which presents serious environmental issues such as high levels of carbon emission, thus the use of sustainable alternatives such as geopolymer concrete (GPC). Nevertheless, GPC has exhibited irregular mechanical strength especially early-age strength that restrains its use to a greater extent. Although nano-silica has been funded in impacting positively into cementitious composites, its impact on GPC is not fully exploited particularly with regard to optimization of dosage and microstructural interactions. To fill this gap research was conducted to study the effect of nano-silica (0 4% by fly ash weight) on mechanical and microstructural properties of GPC. Controlled experimental design was undertaken, whereby compressive, tensile, and flexural strength tests as well as SEM and XRD tests were carried out. ANOVA, Tukey, HSD, and regression were used in statistical assessment. The findings showed that 3% nano-silica produced the maximum compressive strength (39.66 + 0.91 MPa, *p* < 0.001) which was 24.2 percent higher than that of the control whereas tensile and flexural strength were increased by 32.4 and 24.2 percent, respectively. By microstructural examination, more dense matrices with lower porosity (8.5 percent at 3 percent nano-silica as compared to 12.5 percent control) were observed. But the workability decreased in a linear manner as the dosages increased (slump: 81.22 +/- 2.03 mm to 69.01 +/- 1.49 mm). At 2 3%, the dosage provided the best combination between strength improvement and ease of handling, beyond which effects on agglomeration were found. These empirical results support the use of nano-silica in sustainable construction showing its effectiveness in enhancing the performance of GPC. The research can fill the existing severe knowledge gaps in nano-modified GPC, providing practical solutions in material optimization and low-carbon infrastructural construction.

DATA MINING BASED VERTICAL HANDOVER DECISION FRAMEWORK FOR 5G NETWORKS

Rahat Ullah — 2025-08-02

Effective and Seamless execution of vertical handover (VHO) is critical for maintaining sustained connectivity and high Quality of Service (QoS) in 5G heterogeneous networks. However, the differences in network behaviors and protocols make VHO decision-making complicated, often leading to increased latency and service interruption. This paper presents a framework of VHO decision-making using data mining-based techniques within 5G networks. The framework captures historical handover behaviors by applying multivariate regression analysis and Analysis of Variance (ANOVA) to identify significant network parameters like received signal strength, bandwidth, jitter, latency, packet loss and coverage. Through simulations conducted in the NetNeuman environment, it is shown that the proposed framework outperforms the baseline algorithms in terms of enhanced network performance, reduced latency, and improved handover success rates. Real-time decision making based on historical data improves framework responsiveness to user demands, enhancing overall user experience and network dependability. Advanced machine learning systems could be integrated in the future to allow adaptive and predictive mobility management for 6G networks. This research helps in formulating intelligent, data-oriented handover mechanisms required to support ultra-reliable low-latency communications and mobility in next-generation wireless networks.

ANALYZING CARBON DEBRIS AND ENGINE WEAR IN SINGLE CYLINDER DIESEL ENGINE

Faheem Ahmed Solangi — 2025-08-25

In this investigation, three fuel samples—PD100, (2) D96-Bu4 (96%vol. diesel Bu4%vol. N-butanol), and (3) D96-Pn4 (96%vol. diesel, 4%vol. N-pentanol)—were put through endurance test in a single-cylinder CI engine. The results of the study showed that during tests on all gasoline samples, visual inspection showed minor deposits on the engine head. SEM tests revealed that the D96-Bu4 engine had higher carbon deposits on and around the engine head surface than the engine running with DF.Nonetheless, there was less carbon buildup in the ternary mix D96-Pn4. Currently, n-pentanol, diesel, and leftover cooking oil were used to create fuel mixes. When compared to PD, the wear debris concentration was reduced by emulsion fuel in the binary blend, even when n-pentanol was added as a ternary blend D96-Pn4 for aluminum (Al), calcium (Ca), and cadmium (Cd). Ultimately, the viscosity and density readings decreased when the engine was run on both blend fuels.

ANALYZING CARBON DEBRIS AND ENGINE WEAR IN SINGLE CYLINDER DIESEL ENGINE

Faheem Ahmed Solangi — 2025-08-25

ANALYZING CARBON DEBRIS AND ENGINE WEAR IN SINGLE CYLINDER DIESEL ENGINE

Faheem Ahmed Solangi — 2025-08-25

ANALYZING CARBON DEBRIS AND ENGINE WEAR IN SINGLE CYLINDER DIESEL ENGINE

Faheem Ahmed Solangi — 2025-08-25

ANALYZING CARBON DEBRIS AND ENGINE WEAR IN SINGLE CYLINDER DIESEL ENGINE

Faheem Ahmed Solangi — 2025-08-25

ANALYZING CARBON DEBRIS AND ENGINE WEAR IN SINGLE CYLINDER DIESEL ENGINE

Faheem Ahmed Solangi — 2025-08-11

Test Paper Upload

test — 2025-08-05

Abstract here

ANALYZING CARBON DEBRIS AND ENGINE WEAR IN SINGLE CYLINDER DIESEL ENGINE

Faheem Ahmed Solangi — 2025-08-25

ANALYZING CARBON DEBRIS AND ENGINE WEAR IN SINGLE CYLINDER DIESEL ENGINE

Faheem Ahmed Solangi — 2025-08-25

ANALYZING CARBON DEBRIS AND ENGINE WEAR IN SINGLE CYLINDER DIESEL ENGINE

Faheem Ahmed Solangi — 2025-08-25

ANALYZING CARBON DEBRIS AND ENGINE WEAR IN SINGLE CYLINDER DIESEL ENGINE

Faheem Ahmed Solangi — 2025-08-25

INTELLIGENT ASSISTIVE DEVICE FOR VISUALLY IMPAIRED PEOPLE - A COMPUTER VISION BASED APPROACH

Saleem Khan — 2025-08-08

The “Intelligent Assistive Device” system leverages advanced technologies such as Artificial Intelligence (AI), Computer Vision, and the Internet of Things (IoT) to enhance the lives of individuals with visual impairments. This research addresses existing limitations and aims to develop a highly usable, efficient, and cost- effective device tailored to the specific needs of visually impaired individuals. Key features of the proposed system include facilitating independent navigation through obstacle detection, recognition, and distance measuring. Supplementary functions encompass face and currency recognition, wet floor and fall alerts, live location track- ing, text reading assistance, guardian monitoring, and emergency dialing. The functionalities are intended to be incorporated iteratively, empowering blind individuals to perform daily tasks with minimal assistance. The device’s primary features are achieved by integrating IoT and computer vision technologies, utilizing Raspberry Pi, and employing AI-based methodologies. An in-depth analysis of existing assistive technologies informed the identification of their shortcomings and incorporation of their advantages into the design of the proposed system. To ensure user comfort, considerations such as wearability, mobility, and lightweight design have been taken into account. Furthermore, a cost analysis was conducted to develop an affordable yet feature-rich assistive device. The overarching goal of this system is to enhance the quality, productivity, and independence of visually impaired individuals, thereby mitigating the financial and productivity losses associated with visual impairments.

BRINGING AUTONOMY AND COOPERATION TOGETHER: A COMPARISON OF AGENTIC AI SYSTEMS AND AI AGENTS

Muhammad Ahmad Hanif — 2025-08-08

The rapid evolution of artificial intelligence has led to the emergence of two distinct but interdependent paradigms: AI agents and agent-based AI systems. While AI agents focus on modular and task-specific automation, often powered by large language models (LLMs), agentic AI systems represent a conceptual leap by enabling multi-agent collaboration, dynamic reasoning, and persistent autonomy. This article presents a comparative analysis that draws from both theoretical and practical perspectives, integrating the ideas of two fundamental works in the field. We define and differentiate the architectures, interaction models, and design objectives of each paradigm, examining their application in areas such as health, robotics, business automation, and digital ecosystems. The main challenges, such as hallucination, lack of coordination, and accountability, are identified along with mitigation strategies such as ReAct loops, retrieval-augmented generation (RAG), and causal modeling. Furthermore, we analyze the governance, ethical implications, and industry restructuring triggered by agent-based technologies. Our contribution is a unified framework and roadmap that clarifies terminology, aligns capabilities with real-world complexity, and informs the development of robust, transparent, and scalable intelligent systems. This synthesis offers valuable guidance to researchers, policymakers, and industry leaders who are navigating the transition from automated tools to collaborative intelligent agents.

DETECTING PLANT LEAF DISEASES USING CNN MODELS; A COMPARATIVE STUDY

Muhammad Tayyab Rauf — 2025-08-08

The detection of plant diseases through automated systems has gained significant attention in precision agriculture due to its potential to improve crop yield and reduce reliance on manual inspection. This study presents a comprehensive analysis of vegetable disease classification using convolutional neural networks (CNNs). A dataset containing over 20,000 images covering 15 disease categories and healthy classes was utilised, and both a custom CNN model and pre-trained transfer learning architectures were implemented to assess their efficacy in classifying vegetable diseases.

The research involved detailed experimentation with four models: a custom-designed CNN, VGG19, ResNet50, and Xception. The custom CNN demonstrated promising performance, achieving 87.50% accuracy, highlighting that well-structured lightweight models can provide viable solutions in contexts where computational efficiency is paramount. The VGG19 model, leveraging transfer learning, surpassed the custom model with 89.52% accuracy, while ResNet50 emerged as the top performer, achieving 94.86% accuracy, along with high precision, recall, and F1 score, reflecting its strong generalisation and suitability for practical deployment. In contrast, Xception significantly underperformed, illustrating that model architecture choice and fine-tuning play crucial roles in achieving optimal results for plant disease recognition tasks.

The comparative findings underscore the advantages of transfer learning, particularly with deep architectures like ResNet50, for accurate and reliable disease detection. Moreover, the study highlights the potential of the custom CNN as an efficient alternative for resource-constrained environments. The results pave the way for further exploration into hybrid and ensemble approaches, as well as deployment strategies for field-ready disease detection systems. Future work may focus on class-specific error analysis, model optimisation for edge devices, and strategies for addressing class imbalance to further enhance model robustness.

FAKE NEWS IDENTIFICATION AND CLASSIFICATION USING MACHINE LEARNING

Kashif Liaqat — 2025-08-08

A lot of information comes through the social media and people get 70 percent of their news through the social media. It is however also a nest of wickedness that propagates disbelieves and creates fakes. The paper highlights the semantically based identification of a false news to explore and understand the depth of misinformation and draw semantic knowledge to make dynamic decisions. The fake news recognition system targets to formulate an ontology to recognize hypothesis that is employed to swindle social media users by means of logical inference. The given model implies dividing the news content into the fictitious categories and semantically analyzing the news content of the data set. FNIOnt results are projected to three of ML based classifiers to classify the false news: Random Forest (RF) classifiers, Logistic Regression (LR) classifiers, and long short-term memory (LSTM) classifiers. The suggested method is superior to the previous fake news methods, and its identification and accuracy rate is 99 percent. The above findings confirm that machine learning models are better than previous models after the semantic feature investigation on new data sets. The other challenge, which is vital in the task of detecting fake news, is the high rate of adaptation of various strategies by the people behind the identity of fake or misleading news. Since machine learning systems are improving their performance in identifying fake news, maskers of fake stories are constantly changing their methods, either discovering new methods of avoiding detection or changing their writing styles. As an example, they can resort to less direct methods of manipulation like the use of half-truths or statements that are hard to argue with, thereby making the identification more complicated. As a reaction, the machine learning models will have to be made dynamic to respond to these novel methods and can enhance over time by analyzing new data and leaning to the new trends in the generation of fake news. Fake news detection is one of the areas where deep learning, a branch of machine learning using learning with multiple layers that take the form of artificial neural networks, was shown to have potential. Neural networks such as recurrent neural networks (RNNs), convolutional neural networks (CNNs), and transformers can automatically infer complex information in raw text and so the manual selection of features is unnecessary. These models are very applicable in the processing of unstructured text data because they grasp semantic and syntactic relationship between words as opposed to machine learning models, which are limited to understanding such relationships. To illustrate, deep learning models can be trained to approach context and sentiment of a news article to allow them to differentiate between real and fake content even under the conditions of minor manipulation.

ENHANCING THYROID ULTRASOUND DIAGNOSIS WITH A HYBRID CNN AND GRAPH ATTENTION NETWORK

Azeem Mansoor — 2025-08-08

Thyroid diseases, such as hypothyroidism and hyperthyroidism, are prevalent endocrine disorders that significantly impact global health. Early detection is crucial to prevent severe complications, but traditional diagnostic methods often face challenges like delayed results, reliance on human expertise, and limited accessibility in remote areas. This study addresses these limitations by proposing a hybrid deep learning model that combines Convolutional Neural Networks (CNNs) and Graph Attention Networks (GATs) for automated thyroid disease detection using ultrasound images. The proposed model leverages EfficientNet-B4 for spatial feature extraction and GAT layers to analyze relational dependencies between features, enhancing classification accuracy. Trained on the Algeria Ultrasound Images Thyroid Dataset (AUTD), the model achieves an accuracy of 92.48%, precision of 93.94%, recall of 92.48%, and an F1-score of 92.87%, outperforming traditional methods such as Sequential CNN with K-Means clustering (81.5% accuracy). Key innovations include dynamic graph construction for localized feature analysis and robust data augmentation techniques to mitigate class imbalance. The system's performance is ensured by intensive experiments, confusion matrix analysis, and multiclass ROC curves that establish its trustworthiness for clinical deployment. This study contributes to medical AI research by presenting a precise, scalable, and deployable early detection of thyroid disease solution. Future developments can involve investigating more sophisticated attention mechanisms, seamless integration with other clinical data sources.

DEEPFAKE VOICE RECOGNITION: TECHNIQUES, ORGANIZATIONAL RISKS AND ETHICAL IMPLICATIONS

Muhammad Talha Tahir Bajwa — 2025-08-08

Deepfake voice technologies have emerged as a significant advancement in artificial intelligence, particularly within speech synthesis and voice cloning. Using deep learning models such as Generative Adversarial Networks (GANs) and autoencoders, these systems can generate highly realistic synthetic voices that mimic human speech. While beneficial for entertainment and accessibility, deepfake voices also pose major risks in misinformation, identity theft, and cybercrime. This paper explores both the generation techniques and detection strategies for deepfake voices, focusing on neural network–based approaches for voice authentication and synthetic speech recognition. It also highlights the ethical and legal implications of deepfake usage, with emphasis on consent, digital trust, and privacy. By critically analyzing recent trends and proposing a framework for detection, the study aims to support the development of robust defenses against malicious voice manipulation.

MACHINE LEARNING BASED SYSTEM FOR PREDICTING FINGER MOVEMENT OF THE ROBOTIC HAND USING SMART GLOVE

Engr. Nadia Sathio — 2025-08-08

In robot-assisted surgeries, robots are familiar with performing many complicated surgeries with minimal invasiveness and flexibility. This research paper aims to propose a machine learning (ML)-based method for predicting finger movement of the robotic hand. The method utilizes Smart gloves with Light Dependent Resistor (LDR)-based sensors to control Robotic hand-finger movements. The ESP-WROOM-32 microcontroller, connected via Arduino IDE and Jupyter software, records real-time finger movements, including flexion and extension, refined by the microcontroller before real-time integration between the Smart glove and robotic hand. The data generated corresponds to different movements of different fingers involved in multi-learning problems, which deal with scenarios requiring the synchronous prediction or analysis of multiple outputs, such as in multi-output regression. To address this problem, we used the ML algorithm (K-nearest neighbors regressor). This regressor has the inherent property of handling the multiple output regression problem. The regressor used was estimated to predict finger movements concerning Root Mean Square Prediction Error (RMSPE). After implementing this algorithm in real-time integration of the Smart glove and robotic hand, our robotic hand has successfully moved the finger toward the smart glove. The proposed method improves control precision, reduces latency, and improves the user experience, potentially revolutionizing artificial limb control and remote robot operation.

FUZZY INTERFACE-BASED WEED DETECTION SYSTEM USING IMAGE PROCESSING TECHNIQUES FOR SMART AGRICULTURE

Hifza Rani — 2025-08-08

Image processing in detecting weeds is a field that is newly emerging and fast-growing, which can be revolutionary in modern agriculture. The technology helps farmers to recognize and monitor weeds in order to apply specific and effective weed control services. This paper is on the development and implementation of an image-capturing and image-processing system and the design of fuzzy logic on a decision-making platform that determines the suitable dosage level of suitably applied pesticide, together with its application spot concerning agricultural lands.

The natural way of fertilizing in the early days of farming included the use of manure and compost from chickens, cows, and horses. Although these natural methods increased the productivity, now, to keep in line with the rising global food demand, advanced image processing processes are also used in addition to that.

We are using MATLAB as the background processing technique in the field images and identification of grassy weed areas in this work. The fuzzy logic system operates on weed coverage and patch values as well as the usage of membership functions of decision making, one of which involves settling on a rate of application of herbicides at particular areas within the field.

Due to the increase in global population and the exhaustion of natural resources, health-related and sustainable agricultural methods are becoming a focus. As depicted in this paper, the application of image processing technologies can be very crucial in curbing such challenges

A COMPARATIVE ANALYSIS OF E-CIGARETTE AND CONVENTIONAL SMOKING -INDUCED PHYSIOLOGICAL AND HISTOLOGICAL CHANGES IN ALBINO MICE

Tasawar Ahmad — 2025-08-08

The rising use of electronic cigarettes (e-cigarettes) as alternatives to conventional tobacco products has prompted concerns regarding their systemic health impacts. This study aimed to assess and compare the physiological and histopathological effects of e-cigarette (T1) and conventional cigarette (T2) exposure in mice. The study divided adult mice into a control group and two smoking exposure groups, T1 and T2. Research conducted for nine weeks included testing hematological, metabolic, and hormonal parameters, along with histological parameters. In the blood, there was less hemoglobin in both treated groups (6.67±0.356 in T1 and 5.982±0.059 in T2 vs. 6.81±0.09 in controls), but more white blood cells in the T1 group compared to both control and T2 group. The platelet count was also higher in T1 (552.4±5.14) compared to T2 group (325.89±0.26). The testosterone level was higher in T2; however, glucose levels rose in both groups but showed a larger increase in T1. Both e-smoking and conventional smoking exposures influenced estrogen level since it elevated in T1 and diminished in T2. Histological observations showed that both groups of exposed mice had changes in the myocardium, glomerulosclerosis, hepatocellular ballooning, and emphysematous changes. However, the structural problems were worse in conventional smoking (T2 group). Exposure to cigarettes caused continuous weight reduction that affected males to a greater extent among the T1 and T2 groups. Research indicates that both vaping and conventional smoking alter the body; however, e-cigarettes lead to greater disruption of immune functions and hormonal systems than traditional cigarettes and cause more severe structural and blood-related harm

A DEEP LEARNING FRAMEWORK FOR SPACE WEATHER PREDICTION: LEVERAGING TWO-DIMENSIONAL CONVOLUTIONAL NEURAL NETWORK FOR SUNSPOT FORECASTING

Maria Abbas — 2025-08-08

Accurate sunspot activity prediction is crucial for space weather forecasting, as it helps protect space-dependent infrastructure. Cloud computing has significantly advanced deep learning techniques, enabling more precise and efficient forecasting models. This study employs Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), and Two-Dimensional Convolutional Neural Networks (2D-CNN) to enhance sunspot prediction accuracy. Leveraging cloud-based frameworks, the proposed approach improves model scalability, optimizes computational efficiency, and enables real-time forecasting. The dataset consists of time-series records of sunspot activity, making it highly suitable for recurrent neural networks. LSTM and GRU effectively capture sequential dependencies, while optimization techniques, including modified particle swarm optimization and hyperparameter tuning, reduce computational complexity and mitigate overfitting. Experimental results indicate that 2D-CNN achieves the highest accuracy 99.39%, with an F1-score of 98.79%, precision of 99.45%, and recall of 99.33%, demonstrating its superior ability to capture spatial correlations in sunspot data. Furthermore, GRU outperforms LSTM in processing sequential data, achieving higher precision (98.80% vs. 97.81%) and F1-score (96.21% vs. 96.11%). These findings reinforce the effectiveness of deep learning, particularly 2D-CNNs, in sunspot forecasting.

DRIVER DROWSINESS DETECTION SYSTEM BY REAL TIME EYE STATE IDENTIFICATION

Hajra Asif — 2025-08-08

The paper proposes a new architecture which plies eye states of a live video feed and receives at 97 percent accuracy; followed by sending signals at the right time before instances of accidents occur and this is an immense problem in the globe since traffic accidents by fatigued drivers are a huge menace. It is a combined CNN and RNN based system. A comprehensive dataset of 4,760 images, comprising 2,380 closed-eye and 2,380 open-eye images captured under diverse driving conditions, is used to train the model.

AN OPTIMIZED FRAMEWORK OF CYBERSECURITY TECHNIQUES FOR PROTECTING THE PERSONAL INFORMATION OF ACCOUNT HOLDERS IN INTERNET BANKING SYSTEM OF PAKISTAN

Yasir Ali Solangi — 2025-08-08

The rapid innovation in digital technology has revolutionized banking services, enabling automatic financial transactions and transforming customer engagement through Internet banking. This shift has led to reduced operational costs and enhanced customer satisfaction; however, it has also introduced serious vulnerabilities, especially in countries like Pakistan where Internet banking remains a relatively new but growing phenomenon. Fraudsters now exploit sophisticated online techniques, raising the stakes for banks facing internal, external, and regulatory cybersecurity threats. This study employed a quantitative methodology using structured survey responses from 350 account holders across Pakistan to examine these challenges. Through descriptive statistics, reliability testing, and Reliability Analysis, Cronbach’s Alpha, the research validated a four-layer Cyber Defense Framework designed to protect digital financial information. Findings revealed significant gaps in technological awareness, procedural security, and trust in digital transactions, underscoring the urgent need for robust frameworks. Practically, the framework provides actionable insights for financial institutions and regulators supporting more resilient system designs, adaptive cybersecurity strategies, and enhanced legal mechanisms to safeguard users. By aligning technological innovation with strategic security measures, this study contributes a context-sensitive blueprint for strengthening Pakistan’s banking sector against emerging digital threats.

LOW-THD 110 V RMS, 60 HZ PROPORTIONAL-INTEGRAL REGULATED SINGLE-PHASE FULL-BRIDGE INVERTER WITH 10 KHZ SPWM AND LC FILTERING

Faqir Hussain — 2025-08-08

The single-phase full-bridge inverter topology here shown illustrates a robust and efficient means of DC input to a stable 110 V RMS AC output at 60 Hz. Sinusoidal pulse-width modulation (SPWM) with a high-frequency triangular carrier signal (10 kHz) allows for precise control of the four IGBT switches, and thus, the generation of a high-quality AC waveform. The use of an LC low-pass filter (L = 4.06 mH, C = 6.23 µF) further improves the output by filtering out the high-frequency components, thus contributing to the significantly low total harmonic distortion (THD) observed in voltage (--0.22%) and current. A closed-loop control system consisting of a bandwidth-high PI controller (Kₚ = 21, Kᵢ = 0.03155) is tasked with maintaining the stability and quality of the output. Through constant comparison of the filtered output with a 60 Hz reference signal, this controller adjusts the PWM duty cycle dynamically to compensate for variations in load or DC-bus voltage. Such adaptive control allows the inverter to maintain its target output within ±2% of the nominal value, even under severe load transients (±50%) and disturbances in the DC-bus. The fast recovery time, within a fraction of a cycle, is reflective of the system's marvelous dynamic response. Such performance attributes make the inverter design ideal for high-quality, stable AC power sourcing applications, such as renewable energy systems and uninterruptible power supplies (UPS).

DESIGN AND PROTOTYPING OF A LOW-COST, LINKAGE-DRIVEN TWO-FINGER EXOSKELETON FOR HAND REHABILITATION

Saifullah Samo — 2025-08-08

Hand rehabilitation remains a key element in restoring motor functions among individuals affected by neurological injuries such as stroke; robotic-assisted therapy has demonstrated therapeutic effectiveness in prior studies. The practical use of current hand exoskeletons remains restricted due to elevated costs and complex designs; this limitation is more pronounced in healthcare systems operating with reduced financial and technical resources. Existing research lacks a verified system that delivers essential finger mobility through a structure that is both low-cost and simple; few designs can be fabricated using basic materials and tools. The main focus of this investigation was the mechanical development and preliminary evaluation of a hand exoskeleton employing a planar linkage system to guide the motion of the index and middle fingers. A solid model was produced using CAD software; the final device layout was based entirely on this model and ensured accurate component dimensions and assembly alignment. The fabricated prototype utilized laser-cut acrylic linkages; actuation was achieved through standard servo motors; a bevel gear pair delivered the mechanical transmission. The control mechanism was managed using an Arduino microcontroller; the electronics were programmed to control finger trajectories based on predefined flexion-extension angles. This prototype introduced a functional concept of mechanical simplicity; the six-bar linkage system employed only easily available elements assembled into a precise therapeutic motion unit. The complete prototype system weighed close to 100 grams; total expenditure for materials remained under $50 USD; no specialized components were required for construction. Device tests showed controlled finger movements in flexion and extension; the outcomes verified mechanical integrity; actuation reliability and electronic responsiveness were confirmed during performance trials. These findings support the potential of a mechanically feasible and economically accessible device; the demonstrated framework holds value for expanding therapy access in underserved healthcare settings. The study confirms that reliable finger mobilization may be delivered through affordable robotic mechanisms; the approach may improve recovery conditions for patients experiencing hand paralysis or post-stroke motor deficits.

INTEGRATED USE OF BIOFERTILIZERS AND ZINC SULPHATE FOR ENHANCED GROWTH AND PRODUCTIVITY OF WHEAT (TRITICUM AESTIVUM L.)

Shakir Ullah — 2025-08-08

A field experiment was conducted during the Rabi season of 2024 at the Govt Post College, Timergara District, Dir Lower Khyber Pakhtunkhwa, Pakistan, to evaluate the response of biofertilizers and zinc sulphate on the growth and yield of maize (Triticum aestivum). The experiment included treatments with phosphate-solubilising bacteria (PSB), Azotobacter, their combination (PSB + Azotobacter), and zinc sulphate at rates of 20, 25, and 30 kg/ha. The experimental soil was sandy loam in texture, nearly neutral in pH (7.8), and low in organic carbon (0.35%). The results indicated that the combined application of PSB, Azotobacter, and zinc sulphate at 30 kg/ha significantly enhanced the growth and yield parameters of maize. Specifically, it recorded the highest plant height (159.03 cm), plant dry weight (162.70 g/plant), crop growth rate (26.25 g/m²/day), number of cobs per plant (1.8), number of rows per cob (16.8), number of seeds per cob (553.4), 100-seed weight (29.3 g), grain yield (6.5 t/ha), straw yield (12.9 t/ha), and harvest index (33.8%). These improvements may be attributed to the synergistic effect of biofertilizers and micronutrient supplementation. Biofertilizers enhance nutrient availability and uptake, particularly phosphorus and nitrogen, by promoting microbial activity in the rhizosphere. Zinc sulphate contributes to various physiological and enzymatic functions essential for crop development. The findings confirm that integrated nutrient management, using biofertilizers in conjunction with zinc supplementation, is an effective strategy for improving maize productivity while potentially reducing dependency on chemical fertilizers.

TEMPERATURE AND RAINFALL TRENDS IN QUETTA VALLEY, PAKISTAN: A CMIP6-BASED ANALYSIS OF HISTORICAL AND FUTURE CLIMATE DYNAMICS

Fayaz Ahmad Khan — 2025-08-08

The paper explores historic and future climatic patterns of temperature and rainfall in Quetta Valley in Pakistan which is an arid region with a high susceptibility to climate change. As indicated by historical analysis, there exist strong warming patterns in minimum and maximum temperatures with an obvious trend of rising through the period under consideration, along with a small but statistically significant decline of the annual rainfall that is further boosting regional aridity. The ongoing warming is expected to be followed by a further increase in temperature that may reach 8°C in maximum temperatures by the year 2100 according to the high-emission SSP585 scenario. Projections of precipitation indicate uneven patterns that overall have a drier (the potential of lower rainfall than the baseline in SSP585) trend. These results indicate the growing climate fragility of Quetta Valley and the strong need in adaptive practices in water management, agriculture production, and sustainability initiatives.

PREDICTING OPTIMAL LINKS IN COMPLEX HUMAN NETWORKS USING STRUCTURAL PATTERN ANALYSIS

Zulfiqar Ali — 2025-08-08

In human complex networks, link prediction aims to predict when missing, deleted, or future linkages may arise. In this work, we use link prediction methods on five different human interaction networks to find the best prediction method for human complex networks. The techniques utilized are based on similarity-based strategies and are mainly concerned with evaluating each network's similarity scores. Eight algorithms were carefully selected and modified for use in networks relating to humans since they have shown encouraging results in other complicated network contexts. To evaluate the predictive power of the applied techniques, our simulation centers on forecasting links that have been eliminated from the network. The datasets are converted into adjacency matrices and then divided into training and probing sets as part of the technique. The chosen methods are used to calculate similarity scores during a training phase that is followed by rigorous testing. Accuracy metrics are then computed for every dataset. This method makes it easier to do a thorough comparison analysis, which makes it possible to determine which prediction method works best. The author used five different datasets to evaluate the performance of eight different methods. The AUC was the evaluation metric that was employed. According to the findings, the Resource Allocation Index (RAI) performed the best on big and complicated datasets out of all the algorithms.

ANALYZING CARBON DEBRIS AND ENGINE WEAR IN SINGLE CYLINDER DIESEL ENGINE

Faheem Ahmed Solangi — 2025-08-25

ANALYZING CARBON DEBRIS AND ENGINE WEAR IN SINGLE CYLINDER DIESEL ENGINE

Faheem Ahmed Solangi — 2025-08-25

RISK FACTORS OF PREGNANCY LOSS USING MACHINE LEARNING ALGORITHMS

Hijab Fatima — 2025-08-09

Pregnancy loss, also known as spontaneous abortion, is the loss of a fetus before the 20th week of pregnancy. According to the American College of Obstetricians and Gynecologists (ACOG), around 15% to 20% of clinically diagnosed pregnancies result in pregnancy loss. We used cross-sectional data from the Bureau of Statistics Punjab (BSP) to investigate the risk factors for pregnancy loss. we compare the accuracy result of pregnancy loss data using different machine learning algorithms Logistic Regression, KNN, LDA, SVM, NB, RNC, CART, BNB, Passive, ETC to see their performance. After a comparison of the performance of the models, we found the best accuracy of the model KNN as 91%. Algorithms of LR, KNN, LDA, SVM, NB, RNC, CART, BNB, and Passive produced over 80% accuracy. Feature selection and feature importance of 28 variables identified using logistic regression, Decision tree classifier and extra trees classifier that the important features highly affecting the risk of pregnancy are total children ever born and place of delivery

LASER-INDUCED BREAKDOWN SPECTROSCOPY FOR SOIL ANALYSIS: RECENT ADVANCES IN NUTRIENT AND CONTAMINANT DETECTION

Muhammad Rashid — 2025-08-09

Laser-Induced Breakdown Spectroscopy (LIBS) is a new source of distinct role which is now being widely used as an efficient, fast and more versatile method of analysis in soil analysis which has contributed to the best sustainable practice of agriculture. The objective of this review is to provide a comprehensive overview of recent advancements in the application of LIBS for soil analysis, with a particular focus on its role in the detection and quantification of soil nutrients and contaminants. This review aims to highlight how LIBS contributes to improving analytical accuracy, enhancing real-time monitoring capabilities, and supporting sustainable agricultural practices through precise soil characterization. More recent developments have centered on defeating some of the critical drawbacks of LIBS accuracy, including matrix effects, moisture content, and variability of particle size. Optimized experimental procedures, such as spatial confinement, addition of a conductive material and laser-induced fluorescence (LIF) support have shown significant increases in detection limit and precision of the analytical method. The combination of machine learning, deep learning, and chemometric processes continue to optimize LIBS applications by allowing predictive models that can withstand the balkier soil matrices. Moreover, portable and handheld LIBS have contributed to its use in field based real time soil monitoring. Reproducibility is being promised by efforts of standardization through certified reference materials and interlaboratory protocols into increasing acceptance by the scientific community. All these innovations make LIBS one of the most promising instruments in terms of accurate soil nutrient management and contamination testing, with valuable security providing strategic resources, resource-efficient and environmentally sustainable agricultural systems.

EFFECT OF HUMIDITY ON THE DIMENSIIONAL STABILITY OF POLYMER COMPOSITE MATERIALS

Zia Ullah Khan — 2025-08-09

This research analyzes the hygrothermal properties of carbon fiber epoxy composites by investigating the effects of temperature (ranging from 10 C° to 50 C°) and high relative humidity (90%) on the dimensional stability over different durations. A full experimental procedure was done on thirty specimens placed in a climate chamber at the specified temperatures and 90% relative humidity for different time intervals up to 432 hours. A baseline of 14.76 mm Average width and 3.19 mm Average thickness was established.

Dimensional changes included width increase at 10 C° (1.63% to 15.00 mm), 20 C° (1.35 % to 14.96 mm), 40 C° (2.03% to 15.06 mm) and 50 C°(0.27% to 14.80 mm), however, thickness slightly change (0.31%) at 10 C° , 20 C, 50 C° and (0.94%) at 40 C° temperatures. While at 30 C°, width increase (0.068% to 14.77 mm) and the thickness remains the same .The temperature and humidity conditions had clear and profound effects on the mechanical properties of composite materials, thus hygrothermal testing should be used to enhance the durability of composites.

ASYMMETRICAL FOUR U-SLOTS MICRO-STRIP CIRCULAR PATCH ANTENNA FOR WLAN AND WI-FI COMMUNICATION APPLICATIONS

Arshad Wahab — 2025-08-09

The modern communication systems require broad band, compact size and multipurpose antennas for the emerging technologies like 5G and satellite communication. The limitations of micro strip patch antennas include narrow bandwidths and low gains, which make it incompatible with the modern communication systems. The proposed antenna is composed of asymmetrical four U-slots in the patch. The slots are introduced to obtain the broadband characteristics of the antenna. The parameters such as changing the width and length of the U-slots, the radius of the patch and variation in feed point location can improve the efficiency and can remove the bottlenecks of the microstrip patch antennas. The proposed antenna is designed for WLAN and Wi-Fi applications. The aim is to analyze parametric studies and to discuss about how the variations in different parameters of antenna can enhance the gain and the narrow bandwidth. The size and weight of the antenna will be reduced by using an air substrate between the patch and ground.

ENHANCING POWER SYSTEM STABILITY THROUGH THE IMPLEMENTATION OF ADVANCED CONTROL STRATEGIES

Fahiza Fauz — 2025-08-10

This paper examines how advanced control techniques can be implemented on improving stability of a power system with a growing penetration of renewable energy resources, and decentralization of the grid as well as the rising cyber-physical issues. It expects to calculate the effectiveness rates of these strategies, any obstacles in implementation, and the trends in the future regarding the change to resilient and sustainable power systems. The quantitative research design was used where the data was collected via a structured questionnaire to 315 electrical engineering professionals, such as power system engineers (38.1 percent), researchers (23.8 percent), and power system operators (14.3 percent). The responses provided by the participants in relation to challenges, familiarity with strategies, their effectiveness, adoption barriers and technologies of the future were analyzed using descriptive statistics and visual tools (bar charts, donut charts, and frequency tables). The biggest stability challenges were found to be renewable integration (81.0%) and frequency/voltage fluctuations (66.7%). The perceived effectiveness of AI/ML- based control (71.4 percent familiarity) and adaptive control (57.1 percent) was also high (66.7 percent combined Extremely/Very Effective). Major barriers to adoption were low cost (66.7%) and skill gap (57.1). The most significant perceived future technologies were AI/ML (42.9%) and energy storage (28.6%), and 66.7 percent of the results suggested they would probably invest in sophisticated plans in five years’ time. This research paper presents industry professionals opinions on the assessment of advanced control strategies that fill the gap between the theoretical research and practical implementation. It provides implementable policy and stakeholder recommendations to hasten the practicality of adaptive, data-driven solutions that facilitate grid stability within renewable-dominated systems.

Keywords

Power system stability, Advanced control strategies, Renewable energy integration, AI/ML-based control, Grid resilience

GROUNDWATER DEPLETION IN QUETTA: SATELLITE BASED CLIMATE IMPACT ANALYSIS

Fayaz Ahmad Khan — 2025-08-11

Groundwater resources are increasingly vulnerable in arid and semi-arid areas due to rising water demands, climate uncertainty, and unsustainable exploitation. In this study, long-term groundwater depletion in Quetta Valley, Balochistan, Pakistan, is assessed using satellite-derived data extracted from the GRACE and GLDAS missions from 2002 to 2023. The findings identify a constant and extreme decline in terrestrial water storage and groundwater levels, showing seasonal patterns governed by winter precipitation and summer evapotranspiration. Groundwater anomalies obtained from the analysis identify the lack of recovery trends, which highlights the pervasive nature of aquifer stress in the area. The findings are supported by field observations and highlight the urgency of overextraction, which is further intensified by restricted natural recharge and uncontrolled well development activity. The study illustrates the applicability of remote sensing techniques in monitoring groundwater in data-poor regions and calls for an immediate policy response to guarantee sustainable groundwater management in Quetta and similar dryland urban settings.

FACILE LOW-TEMPERATURE SYNTHESIS OF ANATASE TIO₂ NANOPARTICLES AND THEIR APPLICATION IN NANOCRYSTALLINE THIN FILM FABRICATION

Zia Un Nabi — 2025-08-11

Titanium dioxide (TiO₂), particularly in its anatase phase, has emerged as a promising material for applications in photocatalysis, photovoltaics, and transparent conducting films due to its high photoactivity, chemical stability, and non-toxicity. However, conventional synthesis routes often require high-temperature processing (>300 °C), which limits the choice of compatible substrates and increases fabrication costs. This study presents a facile, low-temperature synthesis method for producing phase-pure anatase TiO₂ nanoparticles, followed by their integration into nanocrystalline thin films suitable for optoelectronic and energy applications.

The synthesis approach utilizes a modified sol–gel route at sub-100 °C, incorporating chelating agents and pH control to achieve nanoparticle sizes below 20 nm with narrow distribution and excellent dispersibility. Structural analysis via XRD and TEM confirms the dominance of the anatase phase, while FTIR and UV–Vis spectroscopy reveal functional group coordination and strong optical absorption. Thin films were fabricated through spin-coating and annealed below 200 °C, resulting in uniform, transparent, and well-adhered films.

Film morphology, surface roughness, and porosity were analyzed using AFM and SEM, showing excellent film uniformity and nanostructuring. The optical properties were further evaluated for potential application in dye-sensitized solar cells and photocatalysis, showing enhanced transmittance in the visible region and suitable band gap alignment. Photocatalytic tests using methylene blue degradation confirmed superior activity at low annealing temperatures compared to high-temperature counterparts. The results highlight the strong correlation between synthesis parameters, nanoparticle crystallinity, and film functionality.

This study demonstrates a sustainable and scalable route for anatase TiO₂ nanoparticle production and film fabrication with low thermal budgets. The proposed methodology provides significant advancement toward flexible electronics, transparent coatings, and low-cost solar devices, contributing to environmentally responsible nanomaterial manufacturing.

ENHANCING ENERGY EFFICIENCY IN SMART CITIES THROUGH ELECTRICITY LOAD FORECASTING USING ADVANCED ML MODELS

Moeez Hassan — 2025-08-12

In these rapidly changing times of smart cities, a smart use of energy has become a financial rescue buoy. The forecasting of electricity loads is crucial for the stability of the grid, for resource allocation; however, it also becomes more and more important in the context of integrating wind and solar energy into the grid. Nonetheless, precise prediction is difficult when energy usage changes differently due to the variability of weather, human activity, and renewable generation. Some traditional statistical models, including linear regression and autoregressive type approaches, often fail to model the non-linear and multi-dimensional information underlying the data, resulting in a suboptimal forecasting performance. To address these limitations, this paper applies state-of-the-art machine learning and time series techniques to improve the forecasting accuracy of electricity load. Support Vector Regression (SVR), Random Forest (RF), Gradient Boosting (GB), Long-Short Term Memory (LSTM), the Facebook Prophet, Extreme Gradient Boosting (XGBoost), and Linear Regression are used for prediction. Taking advantage of an extensive electricity consumption dataset as well as time series characteristics and context features, we obtain better forecasting with the proposed model. RF was the best performing among all the models with Lowest MAE=0.021, Extremely low RMSE = 0.014, and Highest R²=0.982 (Almost perfect). These results validate the potential of advanced machine learning based models to provide data capitalism-driven energy management solutions for smart cities.

EMOTIONAL RECOGNITION IN SOCIALLY INTERACTIVE ROBOTS: A COMPREHENSIVE REVIEW

Aryan Ahmed — 2025-08-12

The study explores the evolving landscape of Human-Robot Interaction (HRI) within workplace environments, examining the intricate relationship between humans and robots in professional settings. As artificial intelligence (AI), particularly generative AI, becomes increasingly integrated into the workplace, understanding HRI dynamics is paramount for businesses, society, and ethical considerations. This paper synthesizes current research to provide a comprehensive overview of key trends, challenges, and future research agendas in workplace HRI, examining how advancements in AI, such as natural language processing and adaptive interaction, are shaping human-robot collaboration, communication, and coexistence in professional contexts. The review highlights the growing importance of trust, anthropomorphism, and social intelligence in facilitating effective HRI, drawing upon insights from psychology, computer science, and sociology. It addresses the implications of generative AI technologies like ChatGPT in creating more intuitive and human-like interactions, while acknowledging the ethical dilemmas and potential pitfalls associated with these technologies. Furthermore, the paper discusses how workplace HRI can enhance employee well-being, improve operational efficiency, and foster innovation. By identifying gaps in current research, this review aims to guide future studies, focusing on areas such as long-term user adaptation, AI decision-making interpretability, and the development of robust ethical frameworks for responsible AI deployment in professional environments.

SUSTAINABLE URBAN WATER SUPPLY: A SYSTEM DYNAMICS APPROACH

Zohaib Hassan — 2025-08-12

Water scarcity is an escalating global concern, driven by rapid urbanization, industrial growth, population increase, and climate change. These pressures have led to significant depletion of groundwater resources, particularly in developing countries where approximately 80% of potable water is sourced from underground aquifers. In Pakistan, freshwater infrastructure is often developed in an ad hoc manner, responding to immediate service deficiencies such as low pressure, inadequate quantity, and poor quality at the consumer level. This reactive approach places unsustainable stress on groundwater reserves and results in inefficient financial resource allocation. Despite the urgency, Pakistan lacks a structured framework to guide the planning and development of freshwater supply systems. This study introduces a conceptual systems framework utilizing Causal Loop Diagrams (CLDs) to model the dynamic interdependencies among hydrological, infrastructural, socio-economic, and policy-related factors within urban water supply systems. The framework is intended to support strategic decision making and promote sustainable urban water management. Preliminary validation through expert consultation and scenario analysis highlights its potential to inform resilient infrastructure planning

ARTIFICIAL INTELLIGENCE IN NEURO-ONCOLOGY: INTEGRATING ADVANCED MACHINE LEARNING TECHNIQUES FOR ACCURATE AND EARLY DETECTION OF BRAIN TUMORS THROUGH MRI IMAGING

Ammar Khalil — 2025-08-12

Brain tumors remain one of the most devastating and life-threatening neurological disorders, often associated with high morbidity and mortality rates. Early and accurate diagnosis is critical to improving survival rates and guiding effective treatment strategies. Magnetic Resonance Imaging (MRI) serves as the gold standard for brain tumor visualization due to its superior contrast resolution and non-invasive nature. However, manual interpretation of MRI scans is time-consuming, prone to inter-observer variability, and requires significant clinical expertise, posing challenges in high-volume diagnostic settings. To address these limitations, Artificial Intelligence (AI), particularly Machine Learning (ML) and Deep Learning (DL), is increasingly being applied to neuro-oncology for automated, accurate, and real-time brain tumor detection. This paper presents an in-depth analysis of state-of-the-art AI frameworks designed to enhance the detection, classification, and segmentation of brain tumors using MRI imaging. Advanced algorithms, including Convolutional Neural Networks (CNNs), Support Vector Machines (SVMs), U-Net architectures, and ensemble hybrid models, are evaluated for their ability to differentiate between tumor subtypes such as gliomas, meningiomas, and pituitary tumors. Leveraging open-source databases such as BraTS, REMBRANDT, and Figshare, these models are trained and validated across diverse imaging datasets to assess their robustness and generalization capabilities. Performance metrics such as accuracy, sensitivity, specificity, Dice Similarity Coefficient (DSC), and area under the receiver operating characteristic curve (AUC-ROC) are used to benchmark model effectiveness. The study further explores the benefits of integrating preprocessing techniques like skull stripping, image normalization, and contrast enhancement, which significantly improve model convergence and prediction stability. Additionally, model interpretability and explainability are addressed through visualization tools such as Grad-CAM and saliency maps to support clinical trust and adoption. The paper also highlights the key challenges facing real-world implementation, including data heterogeneity, lack of standardized annotation protocols, limited access to high-quality labeled datasets, and the need for regulatory compliance in medical AI deployment. Ethical concerns, such as algorithmic bias and patient privacy, are critically examined. Overall, the findings demonstrate that AI has the transformative potential to augment clinical decision-making, reduce diagnostic errors, and facilitate timely intervention. With continued advancement and interdisciplinary collaboration, AI-powered MRI analysis is poised to become an indispensable tool in the future of neuro-oncology, offering scalable and precise solutions for brain tumor diagnosis and prognosis.

CLIMATE CHANGE AND ITS INFLUENCE ON THE INDUS RIVER SYSTEM: IMPLICATIONS FOR AGRICULTURAL SUSTAINABILITY IN PAKISTAN

Saddam Hussain — 2025-08-13

The Indus River system serves as the "lifeline" of agriculture for Pakistan, supplying irrigation to around 80 percent of the country's arable land and supporting the livelihood of millions of rural households. The Indus River and its tributaries and distributaries flow in a large network fed by snowmelt and glacial melt from the Himalayas. Because of climate change, this vast system is increasingly threatened by global temperatures rising, altered rainfall patterns, rapid glacier retreat, changing patterns of the monsoon season, and more. All of these climate shifts challenge the variability that farmers have become accustomed to working within and can increase the seasonal variability of river flows, leading to the increased possibility of flooding and drought cycles. Pakistan's food security and economic stability are threatened by climate change to the Indus River system as they are very reliant on predictable water supply from the Indus. Smallholder farmers make up the vast majority of the agricultural sector, and they are dependent on a reliable supply of water while facing variability of supply due to increased variability of supply, decreased crop yields, soil degradation, or potentially shifting growing seasons or cropping patterns. Even as climate change provides us with more evidence of risk, there are still some very large gaps in adaptive water governance, implementation of efficient irrigation, and farmer level climate resilience practices. This paper looks at climate change and the water dynamics of the Indus River, and how these changes impact the future sustainability of agriculture. The paper gives an overview of the complexities of climate change’s impact on agriculture via a mixed methods approach that combines data analysis of long term climate and river flow data, stakeholder interviews and local regional case studies from significant agricultural areas located in the Indus River basin. The analysis separates out regional trends of river flow, identifies vulnerabilities in agriculture, and examines the ability of farming communities to adapt to climate change induced water stresses. The findings of this collaborative research generates information for policy makers, water management authorities, and development practitioners about the pressing need to adopt integrated climate adaptation approaches and efficiencies in irrigation and sustainable land and water management. Therefore, the research also contributes to the literature on addressing agriculture in a climate vulnerable context, with an objective of improving food security for the people of Pakistan and future generations.

AN INFANT CRY INTERPRETER (BABBLE BOT)

Fatima Yaqoob — 2025-08-13

Babble Bot for Infant Communication. Infant crying is a key way by which babies communicate their needs, like hunger, discomfort, tiredness, burping, or pain. Innovating a system that identifies when a baby is crying and identifies the spe- cific reason behind it is included in the goal. The methodologies involved the collection of a dataset combined with algorithms for feature extraction and classification. LSTM model will be used for cry identification and the XGBoost model for reason finding. The cry detection system can tell when a baby is crying and ignore other sounds. Sends real-time alerts to caregivers and works well in different settings. This project is significant because it helps caregivers respond quickly and accurately to the needs of the baby. Knowing when a baby is crying allows caregivers to address things like hunger, discomfort, or pain immediately. This quick response improves baby comfort and reduces caregiver stress by giving them clear information on how to help. The cry detection system will be trained to ignore background noises such as talking, TV sounds, or other ambient sounds and focus only on baby cries. It will work in real time and send instant alerts to parents or caregivers through a mobile app or device. The system is designed to perform well in different environments such as homes, hospitals, or daycare centers. This project is important because it provides caregivers with clear and quick information. Knowing when and why a baby is crying helps them take the right action without delay. This fast response not only improves the comfort of the baby, but also reduces the stress and guesswork for parents, especially new or first-time parents. Furthermore, the Babble Bot can support healthcare staff in neonatal wards, helping to monitor multiple babies at once without missing any cries. Over time, the collected data can also be used for health tracking. For example, frequent pain-related cries might indicate a hidden health problem that needs medical attention. The system can also help detect unusual patterns in a baby’s cries, which may help identify early signs of illness.Babble Bot aims to combine modern artificial intelligence with everyday parenting. It supports responsive care, improves infant well-being, and gives caregivers peace of mind. In the future, this system could be further developed with video monitoring, emotion recognition, or even suggestions for what action to take, making it a complete smart care assistant for infants.

UNVEILING PYTHON-BASED KEYLOGGER MALWARE: BEHAVIORAL ANALYSIS, ARCHITECTURE, AND MITIGATION STRATEGIES

Asad Iqbal — 2025-08-13

The rising sophistication of Python-based malware has made simple scripting languages potent tools for executing surveillance and exfiltration attacks. This paper analyzes a fully operational Python-based Remote Access Tool (RAT) that leverages keylogging, clipboard monitoring, screenshot capture, email-based command-and-control, and self-destruction techniques. Through code-level dissection and architectural modeling, the study reveals the malware’s internal mechanisms and behavior. The paper also proposes detection methods and defensive strategies suitable for individuals and organizations. This research aims to bridge the gap between cybersecurity awareness and technical comprehension, promoting proactive defense against lightweight but dangerous malware.

A COMPARATIVE STUDY OF MACHINE LEARNING ALGORITHMS FOR CARDIOVASCULAR RISK PREDICTION: SUPPORT VECTOR MACHINE, GRADIENT BOOSTING, AND ROTATION FOREST

Muhammad Asad Ullah — 2025-08-14

Cardiovascular heart disease is one of the most fatal problems in the world and is a major cause of deaths globally, reaching around 17.9 million deaths every year. Timely prediction of heart disease is critical for instant response to achieve favorable outcomes; therefore, it requires accurate diagnosis at the right time. Today, the healthcare field has a lot of data, but not much enough knowledge. Machine learning allows computer programs to learn from existing data, get better at doing tasks through experience without needing help from people, and then use what they have learned to make smart choices. There are many different methods and tools in data mining and machine learning that can be used to get useful information from databases and to apply that information for better and more accurate diagnosis. In this research, we compared three machine learning algorithms—Support Vector Machine, Gradient Boosting, and Rotation Forest—to find out which one works best for predicting heart diseases on time. We looked at how accurate each method was, and both Rotation Forest and Gradient Boosting were the most accurate.

DESIGN OF AN ARTIFICIAL INTELLIGENCE-BASED NON-LINEAR CONTROLLER OF A 3-DOF QUARTER ACTIVE VEHICLE SUSPENSION SYSTEM USING AN ELECTRO-HYDRAULIC

Manahil Iqra — 2025-08-15

A vehicle usually encounters road disturbances and an active vehicle suspension system is used to minimize the impact of these road disturbances to ensure a comfortable ride. Vehicles usually comes in contact with sudden road disturbances, such as depression, bumps, acceleration and deceleration, etc and this could lead to an uncomfortable ride and road accidents. So, in this work, a three degree of freedom quarter active vehicular suspension system is modelled and an electro-hydraulic actuator is also modeled for this system. Then an Artificial Intelligence (AI) based non-linear controller is designed for this system, due to which the system becomes more stable, robust and minimizes the uncertainty in the system. A simulation model has been created using MATLAB/Simulink to analyze the impact of road disturbances on a vehicle's suspension system. The results are verified for two different road cases such as no-off-roading case and off-roading case at different speeds and these results shows that the active suspension system model is stable than passive suspension system as the controller is used to improve the system response by stabilizing overshoot, undershoot and settling time of the system.

DESIGN AND INVESTIGATION OF FOUR-SWITCH INTERLEAVED BOOST CONVERTER FOR RENEWABLE INTEGRATION

Moin Khan — 2025-08-15

The primary goal of this research work is to build a DC-to-DC interleaved boost converter having low voltage stress and high voltage gain. The low voltage stress is particularly focused on the switching components, making the converter suitable for low voltage sources like photovoltaic (PV) modules and fuel cells. DC converters, including boost converters, are crucial for reliable performance in various applications. These applications, such as PV systems and fuel cells, often require stepping up the voltage up to 500 V. Conventional converter topologies are not capable of achieving significant voltage gain. High voltage gain of this magnitude cannot be obtained using traditional boost converters, as they require operation at high duty cycles, which leads to issues such as reverse recovery in rectifier diodes. Converters designed with a high voltage conversion ratio tend to suffer from larger voltage drops, resulting in increased switching losses. Furthermore, they also experience losses in parts of the filters which negatively impact the power conversion efficiency. Additionally, smaller filter inductors necessitate minimizing input and output current ripples. This research work aims at obtaining an increased voltage gain while avoiding drawbacks associated with excessively high duty cycles and the reverse recovery issue. The objectives include minimizing the voltage stress on switches, reducing input and output current ripples, and improving overall power conversion efficiency. The study will involve a detailed examination of existing methods, followed by the development of new techniques to meet these goals.

HYBRID ML-BASED FAULT DETECTION IN RENEWABLE-INTEGRATED POWER GRIDS

Engr. Syed Kumail Abbas Zaidi — 2025-08-15

Integration of renewable generation into modern power grids brings new problems in the field of fault detection because now the sources are intermittent and accordingly it is more complex energy mixed. In a network including renewable resources, the conventional fault detection techniques cannot perform to optimum level as they can in constant practice. This work introduces a hybrid machine learning (ML) methodology that synergistically complements the best of both supervised classification algorithms and unsupervised anomaly detection technique to boost fault detection performance. The proposed framework trains a Random Forest, Support Vector Machine and k-Means clustering based ML models with historical operational data that include voltage, current, and frequency measurements. Fusion always makes the decision heavier based on detection robustness by different model outputs. Experiments in simulation and on real-world data show that our approach is more accurate, detects faults faster, and learns to adapt to new grid conditions better than single-model baselines. These findings showcase the efficacy of hybrid AI-based solutions in guaranteeing reliability and resilience within renewable-inclusive power systems.

TEMPERATURE AND RAINFALL TRENDS IN QUETTA VALLEY, PAKISTAN: A CMIP6-BASED ANALYSIS OF HISTORICAL AND FUTURE CLIMATE DYNAMICS

Fayaz Ahmad Khan — 2025-08-15

PERFORMANCE EVALUATION OF LORAWAN UNDER DIVERSE NETWORK SETTINGS: CURRENT APPLICATIONS AND FUTURE PROSPECTS

Muhammad Sohaib Saeed — 2025-08-15

LoRaWAN (Long Range Wide Area Network) is widely recognized as a leading communication protocol for the Internet of Things (IoT). It is designed for long-range, low-power, and low-data-rate wireless communication. LoRaWAN is particularly suitable for Wireless Sensor Networks (WSNs) operating over Wide Area Networks (WANs). These characteristics make it ideal for smart cities, industrial monitoring, precision agriculture, and other large-scale IoT applications. The performance of LoRaWAN largely depends on the Media Access Control (MAC) techniques it employs. Two major MAC techniques are ALOHA and Listen Before Talk (LBT). ALOHA uses random access without carrier sensing. LBT, on the other hand, senses the channel before transmission. Each has advantages and limitations that impact performance under different network conditions. This study evaluates the performance of LoRaWAN and presents a detailed comparison between ALOHA and LBT in the context of LoRaWAN. All the simulations are conducted using OMNeT++ with a network of 100 nodes. The evaluation focused on three key performance metrics: Packet Loss Ratio (PLR), End-to-End Delay, and Power Consumption. The results show that ALOHA is more energy-efficient. It avoids channel sensing, which reduces power usage. However, it leads to higher packet collisions, especially in dense networks. This causes significant packet loss and communication overhead. LBT, in contrast, offers better reliability. It senses the channel before transmitting, which reduces collisions and improves delivery. But this also increases energy consumption due to constant listening. This trade-off is important for designing efficient LoRaWAN-based systems. The choice between ALOHA and LBT should consider the network size, density, and energy constraints. Additionally, this study highlights current and future applications of LoRaWAN. It emphasizes the protocol’s potential as a backbone for next-generation IoT systems that demand scalable, reliable, and energy-aware communication.

SUPPLY CHAIN MANAGEMENT IMPACT ON PROJECT PERFORMANCE IN CONSTRUCTION

Afsheen — 2025-08-16

The construction industry is a major sector that not only delivers essential infrastructure but also ensures safe living environments for communities. This billion-rupee industry is consistently challenged by material shortages, fluctuating costs, logistical disruptions, and labor constraints. In Pakistan, particularly in Karachi, which is the largest urban economic center, there is an urgent need for reliable and efficient supply of materials, manpower, and machinery. This study aims to examine the impact of key supply chain management practices on overall project performance within the construction sector of Karachi. A quantitative research methodology was employed, with data collected from 385 experienced professionals, including project managers, engineers, contractors, and procurement specialists working in prominent construction firms across the city. The data were gathered through structured questionnaires and analyzed using descriptive statistics, correlation analysis, and multiple regression techniques to evaluate the strength and significance of relationships among the identified variables. The results indicate that strategic supplier partnerships exert the most substantial and positive influence on project performance. Additionally, internal lean practices, when effectively integrated with these partnerships, contribute to improved project efficiency and outcomes. The study also identifies several practical challenges that organizations encounter when adopting and implementing lean supply chain practices. This research offers valuable insights for construction professionals, project managers, and decision-makers by emphasizing the importance of developing strong supplier networks, ensuring transparent and timely information sharing, and fostering active client engagement to enhance overall project success.

AN AI-POWERED HIERARCHICAL DEEP NEURAL NETWORK (HIDENN) APPROACH FOR COMPUTATIONAL SCIENCE AND ENGINEERING"

Kinza Urooj — 2025-08-16

In this work, a unified AI-framework named Hierarchical Deep Learning Neural Network (HiDeNN) is proposed to solve challenging computational science and engineering problems with little or no available physics as well as with extreme computational demand. The detailed construction and mathematical elements of HiDeNN are introduced and discussed to show the flexibility of the framework for diverse problems from disparate fields. Three example problems are solved to demonstrate the accuracy, efficiency, and versatility of the framework. The first example is designed to show that HiDeNN is capable of achieving better accuracy than conventional finite element method by learning the optimal nodal positions and capturing the stress concentration with a coarse mesh. The second example applies HiDeNN for multiscale analysis with sub-neural networks at each material point of macroscale. The final example demonstrates how HiDeNN can discover governing dimensionless parameters from experimental data so that a reduced set of input can be used to increase the learning efficiency. We further present a discussion and demonstration of the solution for advanced engineering problems that require state-of-the-art AI approaches and how a general and flexible system, such as HiDeNN-AI framework, can be applied to solve these problems..

A SUSTAINABLE APPROACH TO CONCRETE PRODUCTION USING RECYCLED PAVEMENT MATERIALS

Ali Ajwad — 2025-08-19

The growing deterioration of the environment, fueled by rapid urbanization and excessive resource consumption, has highlighted the urgent need for sustainable construction practices. One promising approach is the incorporation of recycled pavement waste aggregate as a partial replacement for natural aggregate in concrete production. This study investigates the environmental and structural viability of using recycled aggregate in concrete by replacing 5%, 10%, 15%, and 20% of conventional aggregate. The workability of fresh concrete was evaluated through slump cone and compaction factor tests, while compressive strength tests were conducted to assess the performance of hardened concrete. Experimental results reveal that concrete containing recycled aggregate demonstrated satisfactory performance compared to conventional concrete. The findings suggest that partial replacement of natural aggregate with recycled material can reduce construction costs, conserve natural resources, minimize energy usage, and offer an environmentally responsible alternative without significantly compromising strength and durability.

SMART BLIND KIT: ARDUINO-BASED WEARABLE ASSISTIVE DEVICE FOR REAL-TIME OBSTACLE DETECTION AND MOBILITY SUPPORT

Muhammad Bilal Alam — 2025-08-19

The demand of affordable and efficient assistive technology has increased with the increasing number of the sight impaired people across the world. The current research introduces the design and construction of Smart Blind Kit, a low-cost sensor-based portable navigation aid designed in open-source hardware and a combination of low-cost parts. The system uses Arduino Uno microcontroller together with ultrasonic and infrared sensors, piezo buzzer, and RF-based locator module in order to offer real-time obstacle detection using vibration module and audio-based feedback. The modular architecture comprises sensor-on-sandal to sense the ground level obstacles and an IR-sensor band on the head to identify the hazards in the lateral and upper sides. A complete design of a hardware prototype was developed and field-tested. Experiments showed consistent performance in detecting the obstacles with a very low latency and a comfortable and easy-to-learn style of feedback. With the number of six ultrasonic sensors constructed on the shape of a headband, the ultrasonic circuit can scan a large area of field of view, whereas the sandal circuit linking numerous IR sensors and one ultrasonic sensor can offer an extended scope of space. The breadboards were used to build both circuit schematics and to validate it, which is included in the contribution. The usability, accuracy, and the power consumption of the system were checked. It was found to be useful in indoor and semi-outdoors areas where test users gave positive reviews to its use. The fact that the device consumes low power, has an ergonomics of wearable and is scalable supplements the fact that the device will be ideal to be deployed in developing countries. This project provides a scalable backdrop to the future additions of GPS navigation, smartphone interconnectivity, and voice control and is a potentially value-adding solution to facilitate safe and independent movement in visually impaired persons.

BUILDING TRANSPARENT PROFESSIONAL REPUTATION: A BLOCKCHAIN FRAMEWORK A TRUSTWORTHY BADGE BASED IDENTITY SYSTEM

Marrium Asif — 2025-08-19

In today's advanced digital world, it is very important to construct a professional reputation on which people can trust. This paper will recommend a badge identification system which is based on Blockchain that gives clear-cut, unbreakable proof of credentials and professional achievements. In this proposed approach, it will incorporate Smart Contracts and Decentralized Identifiers (DIDs) to provide trust, scalability, and privacy to the people. A Blockchain with limited access will be established so that companies can issue and give authentic digital badges which is suitable for reputation verification, and use some security analysis to evaluate the system. The results show that badge systems that are using blockchain are more safe and scalable alternative to traditional ways of checking credentials.

This study represents a blockchain-based system for badge-driven identity and reputation management which is used for Verifiable Credentials (VCs), Smart Contracts and decentralized identifiers (DIDs) to get rid of the problems which are faced in traditional way. In this suggested system, recognized institutions may provide professionals credentials that are temper-proof and can be verified by cryptography. These credentials may be used as proof of affiliation. These badges are linked with the self-governing identities, so that users can be sure about their work.

A reputation score model will be constructed by putting together endorsement graphs, badge information, and issuer credibility that will see how trustworthy a professional is working. The system uses Ethereum smart contracts, open-source DID protocols to create the identification without central authority, and IPFS which will store badges off-chain.

The research will contribute to the domains of decentralized identity management, blockchain applications in a trusted system, and algorithmic reputation modeling in both theory and practice. The suggested system will make the possibility to check and transfer professional credentials, that will open up new possibilities for education, job verification, freelancing platforms, and decentralized learning ecosystems. At the end, it gets closer to the aim of a digital reputation system that users can manage and trust, which is important for the new Web3 and knowledge economy.

HIERARCHICAL ATTENTION SPATIOTEMPORAL GRAPH NEURAL NETWORK WITH DYNAMIC MODALITY WEIGHTING FOR STAGE-SPECIFIC PARKINSON’S DISEASE DETECTION

Amna Arshad — 2025-08-19

Parkinson’s disease (PD), a progressive neurodegenerative disorder, affects over 10 million people globally, necessitating early detection to enable timely interventions that enhance quality of life. Current diagnostic methods, such as the MDS-UPDRS, rely on subjective clinical assessments, often missing subtle early-stage symptoms like minor gait changes or hand tremors. This paper proposes a Hierarchical Attention Spatiotemporal Graph Neural Network with Dynamic Modality Weighting (HAST-GNN-DMW) for stage-specific PD detection using multimodal computer vision. Integrating gait, hand movement, and speech data from the PPMI, PD-Posture-Gait, and a synthetic PD-MultiStage dataset (300 patients, Hoehn-Yahr labeled), our framework employs hierarchical attention to model intra- and inter-modality dependencies and dynamically weights modalities based on patient-specific symptom severity. Explainable AI (XAI) via Integrated Gradients identifies key biomarkers, such as stride length and tremor frequency, enhancing clinical interpretability. Evaluated on PD-MultiStage, HAST-GNN-DMW achieves 93.8% accuracy in early PD detection and 90.5% in stage classification, outperforming state-of-the-art methods like ST-GCN and DenseNet. Ethical protocols ensure fairness through balanced datasets and GDPR-compliant anonymization. Limitations include dataset size and real-world noise sensitivity, with future work targeting larger cohorts and edge-based telemedicine deployment. This framework offers a scalable, interpretable solution for early PD diagnosis, advancing clinical adoption and improving patient outcomes.

A HYBRID DEMAND FORECASTING AND REINFORCEMENT LEARNING FRAMEWORK FOR DYNAMIC PRICING IN E-COMMERCE

Kashif Karim — 2025-08-19

Dynamic pricing is becoming a hot topic these days, most importantly in e-commerce to increase the number of sales and customer satisfaction. This technique is mostly used in developed countries but in developing countries like Pakistan the applications of dynamic pricing are restricted. The main purpose of this research is to scrutinize the implementation of a hybrid dynamic pricing model in developing countries like Pakistan e-commerce sectors by the integration of reinforced learning (RL), demand forecasting, and price prediction methodologies. This study employs Reinforcement Q-Learning and Deep Q-Learning to simulate real-time pricing scenarios. Furthermore, in this study demand forecasting is performed using ARIMA and Prophet models, while Random Forest and XGBoost algorithms are implemented for accurate price prediction based on product-level features. The goal is to create a dynamic pricing structure that is fair, adaptable, and appropriate for the particulars of Pakistan's online marketplace. Dynamic pricing can bring big changes in online marketplace in developing countries. However, its success depends equally on the social and economic conditions as it relies on advances in technology. The ultimate goal of this study is to create a sustainable pricing model that takes into account the preferences of Pakistani consumers and, using data-driven insights and adaptive learning techniques, adapts product prices to shifting market conditions.

BANK REPAYMENT PREDICTION-SYSTEM ON DEEP LEARNING TECHNIQUES

Rafique Ahmed Vighio — 2025-08-20

A significant service provided by financial institutions is loans, however, with little resources to spread out banks should be selective in selecting low-risk borrowers in order to reduce the number of defaults. The traditional machine learning algorithms such as SVM and KNN are limited in one way or another; SVM is too computationally intensive, KNN tends to perform significantly worse because of inefficiency and parameter sensitivities. Similarly, Random Forest, despite its popularity, suffers in two aspects, namely, feature selection and model tuning.

This research paper proposes a deep learning-based system to predict bank loan repayment to address such shortfalls. Through an instance of Deep Neural Network (DNN), the model can learn and optimize features automatically, can support more kinds of data, and achieves a higher degree of accuracy and scalability than the classical methodologies.

A dataset of 100,000 loan applications provided by Kaggle was used to train the proposed model on. Its accuracy of 82% was better than the 68% exhibited by Random Forest by 13%. The results indicated the possibility of deep learning to minimize the bad loans, enhance the risk assessment, and also to optimize the future loan approvals.

KeywordsDeepLearning,LoanPredictionModel,raining,Testing,Prediction,AccuracyAnalysis.

UNCOVERING UNDERGRADUATE BEHAVIORAL PATTERNS: A STUDY OF ACADEMIC AND SOCIAL PROFILES ACROSS FOUR UNIVERSITIES USING CHI-SQUARE AND K-MODES CLUSTERING

Salman Ahmad — 2025-08-21

Quantitative understanding of relationships between students’ behavioral patterns and academic performances is a significant step towards personalized education. This study investigates behavioral patterns among undergraduate students across four universities, focusing on various behaviors like hours spent on social media, internet consultation during academic tasks, CGPA, daily study hours, preferred entertainment, and preferred mode of social interaction. Data were collected through a structured questionnaire comprising twelve categorical variables. Chi-square tests identified several significant associations, including the relationship between CGPA and daily study hours, as well as between preferred entertainment type and social interaction mode, indicating clear behavioral dependencies. K-modes clustering, with the optimal number of clusters determined via the elbow method, yielded five distinct student profiles. Cluster profiling revealed patterns such as academically focused students with limited leisure activities, socially active students with moderate academic engagement, and balanced profiles combining both strong academic performance and diverse leisure interests. These findings offer valuable insights into how behavioral tendencies correlate with academic outcomes, contributing to targeted educational strategies that address both performance enhancement and student well-being.

DNN-BASED INTRUSION DETECTION FOR ENHANCING LORAWAN SECURITY: CURRENT APPLICATIONS AND FUTURE PROSPECTS

Muhammad Sohaib Saeed — 2025-08-21

The rapid adoption of LoRaWAN technology in smart cities, industrial IoT, and critical infrastructure has amplified concerns over its vulnerability to sophisticated cyber-attacks. Existing security solutions often struggle to meet the accuracy, adaptability, and scalability requirements of resource-constrained, long-range communication systems. This paper proposes a Deep Neural Network (DNN)-based Intrusion Detection System (IDS) for Long Range Wide Area Network (LoRaWAN)-based smart communications. The model architecture comprises multiple fully connected layers with Rectified Linear Unit (ReLU) activation for effective non-linear feature extraction. The Adam optimizer is employed to achieve accelerated convergence during training, and a Softmax output layer is used to perform multi-class classification across the different attack categories. The CICIDS2017 dataset, a comprehensive and realistic benchmark for network intrusion detection research, is used for performance evaluation. Experimental results reveal exceptional detection capability against complex cyber-attacks under a vast range of performance parameters. The proposed IDS have achieved 99.98% accuracy, 99.99% precision, 99.99% recall, and 99.99% F1-score. These outcomes demonstrate the model’s robustness in differentiating legitimate LoRaWAN traffic from a wide spectrum of malicious activities in real time. Furthermore, the proposed approach exhibits high generalization potential, making it suitable for deployment in diverse Internet of Things (IoT)-based environments. Future work will focus on lightweight model optimization and real-world LoRaWAN traffic validation to ensure practical applicability in large-scale smart communication networks. Additionally, this research underscores the current applications along with the future prospects of secure LoRaWAN communications.

ANALYZE THE IMPACT OF ARTIFICIAL INTELLIGENCE, NETWORK ARCHITECTURE, AND HUMAN ERROR ON CYBERSECURITY THREAT DETECTION

Aysha Ijaz Khan — 2025-08-21

Sometimes, conventional threat detection systems are inadequate given the growing complexity and frequency of cyber-attacks. One strong approach to improve threat detection skills has emerged from the incorporation of Artificial Intelligence (AI). Network design and the continuing part of human error in cyber security breaches, however, affect the efficacy of artificial intelligence.This study seeks to investigate how human elements, various network architectures, and artificial intelligence-based systems collectively affect the efficacy and precision of cyber security threat detection. A cross sections study design was carried out, using simulated corporate network environments across three architectures: flat, segmented, and zero trust. AI algorithms including machine learning-based anomaly detection and behavior analysis tools were used throughout these settings, With real-time logging of security incidents and response accuracy, data were gathered six months. Human error data were collected using structured incident reporting forms from 100 cyber security professionals aged 25–55 years. Stratified random sampling was used to ensure representativeness across organizational types. Statistical analysis using logistic regression and ANOVA evaluated the individual and cumulative influences of human error, network architecture, and AI performance on threat detection rates. In zero trust topologies, AI-enhanced systems achieved an average detection accuracy of 94.6%; in segmented networks, 88.3%; and in flat networks, 72.5%.Statistical analysis verified that both network architecture and human error significantly affected AI performance (p < 0.01). Although AI greatly enhances cyber security threat detection, its efficiency depends much on strong network infrastructure and diminished human error. Forty-one percent of missed or delayed threat responses were attributed to human mistake, mostly resulting from alert misinterpretation or delayed escalating. To attain resilient cyber security, organizations need to use whole security approaches combining best network design, advanced artificial intelligence tools, and extensive human training.

PRACTICAL APPROACH: IOT BASED CROP IRRIGATION MANAGEMENT SYSTEM

Muhammad Amir — 2025-08-21

Agricultural yield is an essential part of any country's economy. More so for Pakistan, where over 70% of the population is connected to a farming-based economy, while 25% of the nation's GDP derives from farming-based activities, playing a crucial role in the country's growth. Various issues related to the field of agriculture continually hamper such growth; lack of real-time monitoring and management of crops is one of them. An inevitable solution to such problems is to opt for modernized farming techniques and incorporate the Internet of Things (IoT) monitoring adopted by the developed world. Using IoT-based real-time monitoring, crop management can become smart with enhanced yield. Smart plantation raises crop production and reduces wastage. The central aspect of this study is to provide an accessible and cost-effective solution for real-time monitoring and control of essential parameters, such as timely irrigation, that affect agricultural yield. Such tracking, in real-time, initially involves transferring all real-time sensor(s) data to the cloud for processing. Secondly, the processed real-time data will be available on an Android smartphone App, providing farmers with quick IoT-based situational awareness & control of an intelligent irrigation system with optimized water usage.

OPTIMIZING FVD ORIENTATIONS FOR TALL RC BUILDINGS UNDER COMBINED SEISMIC & WIND HAZARDS

Umer Shahzad — 2025-08-21

A 20-storey reinforced-concrete frame (designed per ACI 318-19, ASCE 7-16 and local BCP-21 codes) is analyzed under nonlinear time-history seismic loads (using spectrally matched ground motions) and ASCE wind loading for a Karachi like coast. Three supplemental fluid viscous damper (FVD) layouts are studied: single diagonal braces, chevron bracing, and wall mounted dampers. Compared to the bare frame, all damper configurations yield substantial improvements in dynamic response. Under the design seismic sequences, the wall-mounted dampers provide the strongest control: peak inter-storey drift is cut by about 42% (to ~0.00618) and roof displacement by roughly 50%, while ~58.6% of the input seismic energy is dissipated by the system. The chevron arrangement also significantly reduces response, absorbing about 42.1% of the energy, and the single diagonal layout about 36.8%. In absolute terms these devices roughly halve the undamped building’s drifts and displacements across height. Under wind excitation, the trends differ: chevron or diagonal orientations better suppress low-frequency sway, whereas the wall-mounted scheme maximizes energy dissipation and drift control under earthquake shaking. These results highlight a tradeoff in mixed hazards. For a high-wind, high-seismic site like coastal Karachi, a hybrid strategy is advised: e.g., chevron dampers on the windward facade to counter aerodynamic sway, combined with dense wall mounted dampers on the transverse axis for optimal seismic energy dissipation. Such damper layouts provide actionable guidance for resilient tall building design under combined seismic and wind loads.

Keywords ( Fluid viscous dampers; reinforced-concrete frame; nonlinear time-history analysis; inter-storey drift; seismic energy dissipation; wind-seismic interaction)

A SECURITY-CENTRIC ARCHITECTURE FOR BIG DATA

Shaukat Ali — 2025-08-21

Big Data is a hot area of research and a buzzword in the research community and industry. The term Big Data refers to the huge volume of data, having properties of high production of even more data (velocity) with different forms of structured and unstructured data (Variety). Big data also has big problems associated with it; one of these problems is the issue related to security and privacy. In this paper, a security-centric architecture is proposed for big data. The proposed architecture protects data at the application layer using a granular access control mechanism and secures the data under management at the storage layer using a data aggregation method. The proposed architecture has a layered approach. This architecture provides a base for the community of big data security. The paper also explains different layers of architecture, especially those related to data security and privacy issues.

IMPACT OF MACHINE LEARNING ALGORITHM CHOICE AND DATA QUALITY ON MODEL ACCURACY

Nisha Rafique — 2025-08-22

This study investigates the impact of machine learning (ML) algorithm choice and data quality on model accuracy. With the growing adoption of ML across industries such as healthcare, finance, and environmental sciences, understanding how different algorithms perform under varied data conditions is essential for optimizing model performance. The study examines five widely-used ML algorithms—Decision Tree, Random Forest, Support Vector Machine (SVM), Neural Network, and Gradient Boosting—across five publicly available datasets manipulated to simulate high and low-quality data conditions. Statistical analyses, including One-Way ANOVA, Independent Samples t-test, and Two-Way ANOVA, reveal that both algorithm choice and data quality significantly influence model accuracy. The results indicate that ensemble methods like Random Forest and Gradient Boosting are more robust to poor-quality data compared to simpler models such as SVM and Decision Trees. The study emphasizes the need for careful algorithm selection and data quality improvement in machine learning model optimization, highlighting the critical role of data preprocessing.