Spectrum of Engineering Sciences

DESIGN & IMPLEMENTATION OF SMART GRID USING CELLULAR NETWORKS AS ITS AMI BACKBONE

Abdul Rehman — 2025-05-02

To solve the limitations of current power grid, Smart Grid is the solution. Its backbone is AMI (Advanced Metering Infrastructure), but its major problem is lack of unified data transmission infrastructure within the grids which should able to send big amount of data collected from each smart meter of a town or city to the center in cheap and secure way. In the current design of AMI in SG needs the implementation of a separate/mix transmission data infrastructure which will have very high cost. So, we introduce the use of Cellular networks in transmitting consumer data (usage and controlling) to the center and vice versa. We have designed a system where smart meters equipped with Cellular SIM Cards and Wi- Fi technologies communicate directly with the Hub in a secure way, bypassing the need for extensive infrastructure. The Wi-Fi in smart meters will be used in IoT of customer's home equipment management. This system could be deployed in the whole country just by installing smart meters and renting Cellular Companies Transmission infrastructure to send each customer data to the Hub. This design has been simulated in MATLAB. So, this approach associated with low-cost, high security and easy maintenance as the Cellular network's companies will be responsible for its maintenance just like their new cellular sites. This re-search provides a blueprint for utility companies and policymakers to upgrade electricity systems economically and efficiently using/renting already installed Cellular infrastructure.

INNOVATIVE ARCHITECTURE AND INDUSTRIALIZATION BUILDING SYSTEM IN THE CONSTRUCTION INDUSTRY OF PAKISTAN

Sania Rehman Memon — 2025-05-02

Industrialized Building System (IBS) construction has emerged as a sustainable method for enhancing productivity and mitigating the adverse environmental and social impacts associated with conventional construction practices. While a growing body of research has addressed management issues in prefabricated construction, a systematic summary and strategic framework for its optimization remain underdeveloped. This study aims to develop a framework for optimizing the use of industrialized building construction by integrating innovative architectural approaches and construction waste reduction strategies. Through a comprehensive review of best practices in prefabrication implementation, including adoption rates, construction methods, cost-effectiveness, and performance evaluation, this paper examines effective implementation strategies, industry prospects, and the enabling environment for technological application. Special focus is placed on design, production, transportation, and assembly processes. The findings offer an improved understanding of the critical factors influencing the success of IBS and provide practical guidance for enhancing future adoption, promoting sustainability, and advancing innovation within the construction industry.

HIGH-PERFORMANCE LINEAR TO LINEAR POLARIZATION CONVERTER EXHIBITING 30° ANGULAR STABILITY

Haneef Hamza — 2025-05-02

A thin linear polarization converting metasurface has been proposed with better efficiency and performance. The proposed surface is etched on 3mm thick dielectric substrate. It has the ability to convert linear to orthogonal equivalent in the 10.5-21.5GHz band with polarization conversion ratio above 90%.along with this, parametric analyses has also been conduct to monitor the effect of various parameter on the performance of surface. Moreover, for deeper insight into polarization conversion mechanism the surface current distribution is carried out.

ENHANCED BRAIN TUMOR DETECTION IN MRI: A COMPARATIVE STUDY OF MACHINE LEARNING MODELS

Khalid Mahboob — 2025-05-02

Image processing is essential and attractive in the medical and healthcare. Digital image processing identifies diverse pathological methods, like identifying, classifying, evaluating, and testing brain tumors through microscopic images. Many machine-learning methods are recognized in the era of the AI century for detecting brain tumors through Magnetic Resonance Imaging (MRI). MRI is a recognized image processing method through three-dimensional examination, which identifies unambiguous images of the infection or tumor. The paper aims to offer supervised machine-learning algorithms for brain tumor detection in MRI images through a comparative analysis of different models. Considering the specific features of the tumor and surrounding infected tissues of the brain through analysis supports us in estimating the accuracy of the models and recognizing the optimal operative method. In this paper, four supervised machine learning models are considered: Logistic Regression (LR), Neural Network (NN), Stochastic Gradient Descent (SGD), and Support Vector Machines (SVM). MRI images can quickly identify brain tumors or infections by comparing these models. Furthermore, a model is developed using the Visual Geometry Group (VGG-19) embedder and the Kaggle dataset. The result section shows that the proposed model outperforms the benchmark schemes by attaining high proximity accuracies.

ANDROID MALWARE ANALYSIS USING ARTIFICIAL INTELLIGENCE

Ali Ahmed — 2025-05-03

Mobile phones have become a crucial part of society and serve as more than just communication devices. The growing use of smartphones has led to a large number of apps, making it difficult for app marketplaces to validate their legitimacy. Conventional security solutions for computer malware are challenging to apply on mobile devices due to different resource management mechanisms. Implementing intelligent tools using the Machine Learning in the threat identification process of security software can improve its efficiency by analyzing data and identifying potential threats. This reduces the need for human intervention and allows for faster detection of risks, saving time and resources. Intelligent tools can also continuously monitor data and identify potential threats in real-time, further improving the threat identification process. In conclusion, the use of intelligent tools can significantly enhance the effectiveness of conventional security software and protect against potential threats. This can help prevent hacking and data theft and keep personal information safe and secure. Additionally, these intelligent tools can be easily integrated into current security systems, making it easy for organizations to improve their overall security posture.

SMART FILTERS FOR SMS SPAM: A MACHINE LEARNING APPROACH TO SMS CLASSIFICATION

Ishrat Nawaz — 2025-05-03

The exponential rise in the number of undesired text messages delivered via SMS has been directly related to the explosion in the number of mobile phones sold. Although various information channels are considered "spotless" and trustworthy in many parts of the world, ongoing reports show that cell phone spam is significantly increasing. It is a big problem. It is becoming increasingly pervasive worldwide, especially in Asia and the Middle East. In the same way that finding a solution to such an issue can be time-consuming, so can the process of identifying spam texts from genuine communications. It solves many difficulties and makes life much easier because it can distinguish between real SMS and spam. In any event, it faces specific challenges and obstacles that are unique to itself. During this current research, we have investigated five Machine Learning (ML) methods to identify spam in a short text message using a single dataset containing SMS spam Collection. The SMS spam dataset was extracted from the Kaggle repository. The experiment is carried out on the R platform. Eleven characteristics, including binary and numeric features like Char Count, Has number, Has URL, Has Date, Has dollar, Emoticon, Email, and Phone, as well as spam count, ham count, and spam binary, are employed in this research. These features are used for feature selection and showing results using Machine Learning(ML) approaches. The effectiveness of the various strategies or methods is evaluated using metrics such as sensitivity, accuracy, precision, F1 score, recall, and specificity. The outcomes show that the light gradient boosting machine (LGBM) with these features achieved a sensitivity score of 100, precision score of 100, F1 score of 100, recall of 100, and specificity score of 100, with an optimal accuracy score of 100 percent, which is outstanding compared to all other state-of-the-art studies.

ASSESSING THE AGRONOMIC AND ECONOMIC VIABILITY OF DISTILLERY SPENT WASH AND BOILER ASH MIXTURES AS SUSTAINABLE FERTILIZERS

Muhammad Ali Keerio — 2025-05-03

A significant by-product of the sugar industry is molasses, which distilleries utilize for alcohol production through fermentation. This process generates 10 to 15 cubic meters of wastewater for every cubic meter of alcohol produced. Common disposal methods for this waste effluent include fertilization, irrigation, composting with bio-waste, combustion, and anaerobic treatment. Rich in organic matter, mineral salts, and essential nutrients, this effluent can be efficiently repurposed as a nutrient source and soil amendment. By adopting these methods, industries can enhance soil fertility while managing waste responsibly, contributing to sustainable agricultural practices. Research studies have demonstrated that the nutrient content (NPK) in distillery sludge makes it suitable for use as a fertilizer. A developed fertilizer combining distillery spent wash (DSW) and boiler ash (BA) was tested on cotton fields, resulting in improved crop growth, higher yields, and reduced costs. This approach enhances nutrient availability and offers a viable alternative to synthetic fertilizers for agricultural productivity. A parametric study was conducted to evaluate the effectiveness of this low-cost method, focusing on the utilization of waste materials like DSW and BA from the alcohol industry as sustainable fertilizers. The findings highlight the potential of these waste-derived fertilizers to improve agricultural outcomes while addressing environmental and economic challenges, paving the way for more sustainable industrial and farming practices.

SMART SURVEILLANCE: A REVIEW OF RECENT ADVANCEMENTS, APPLICATIONS, AND ETHICAL CONSIDERATIONS

2025-05-04

Smart surveillance systems are rapidly evolving due to advancements in artificial intelligence (AI), the Internet of Things (IoT), and edge computing. This review examines the recent progress in smart surveillance technologies, focusing on their applications in various domains, including public safety, smart cities, and healthcare. The review also addresses the ethical considerations and challenges associated with the deployment of these systems, such as privacy concerns and algorithmic bias. Furthermore, it identifies future research directions and opportunities for innovation in the field.

A COMPREHENSIVE REVIEW OF CNN ARCHITECTURES FOR IMAGE RECOGNITION: ADVANCES AND OPEN CHALLENGES

Sohail Ahmed Memon — 2025-05-05

Convolutional Neural Networks (CNNs), considered revolutionary, have transformed the field of computer vision, aiding exceptional advancements in image recognition tasks. With their ability to automatically learn spatial hierarchies of features, CNNs have become the backbone of most innovative image recognition systems. From their inception with LeNet in the late 1990s to the latest revolutions like Vision Transformers (ViTs), CNN architectures have undergone significant evolution. This paper provides a comprehensive review of the key developments in CNN architectures, focusing on their impact on image recognition performance, the challenges they face, and the potential future directions.

SECURING IOT DEVICES IN HEALTHCARE: CHALLENGES AND SOLUTIONS

Azeem Akram — 2025-05-06

This review article investigates the use of blockchain technology and powerful machine learning algorithms to improve IoT device security in healthcare. The main objective is to tackle crucial security issues like anomaly detection, authentication, and data integrity. A decentralized, unchangeable ledger for transactions and data transfers is made possible by blockchain technology, guaranteeing strong data integrity and transparency. By detecting abnormalities from typical behavior patterns, advanced machine learning models such as anomaly detection algorithms are used to detect and mitigate security threats in real-time. Furthermore, context-aware dynamic Bayesian networks and blockchain greatly enhance authentication methods, guaranteeing that only authorized users have access to IoT devices and data. The review emphasizes how important it is to strike a balance between strict adherence to regulations and thorough security measures in IoT systems for healthcare. Best practices for defending IoT networks against changing threats are also presented. The results highlight how blockchain and machine learning can be used to secure Internet of Things applications in the healthcare industry, laying the groundwork for more research in this important area.

FROM DETECTION TO PRECISION: ELEVATING HATE SPEECH CLASSIFICATION WITH CUTTING-EDGE MODELS

Nimra Maqbool — 2025-05-06

In the digital era, the proliferation of hate speech on social media platforms has necessitated the development of effective detection systems. This paper presents a comprehensive comparative analysis of machine learning and deep learning approaches for hate speech classification across diverse datasets, including a thorough comparison with existing methodologies. Specifically, this study evaluates the performance of two machine learning models Random Forest and XGBoost and two deep learning models, LSTM and BERT. Each model is trained using various embeddings, including Word2Vec, as well as GloVe, supplemented by TF-IDF for the machine learning models. Through rigorous cross-validation and hyperparameter tuning, the efficacy of each model and embedding combination is assessed. The results are analyzed not only to determine the most effective approach for hate speech detection but also to benchmark these results against previous studies in the field. This comparative analysis provides insights into the strengths and limitations of the models and embeddings used, aiming to contribute to the ongoing efforts in creating a safer online environment by advancing the state-of-the-art in hate speech detection.

ARTIFICIAL NEURAL NETWORKS FOR IMPROVED IMAGE RECONSTRUCTION IN ELECTRICAL IMPEDANCE TOMOGRAPHY

Fakhar Anjam — 2025-05-06

The research applies the Artificial Neural Network (ANN) method to create images by utilizing Electrical Impedance Tomography (EIT) as its image reconstruction system. The ANN technique demonstrates versatility because its applications extend across different domains, which include classification functions with additional enhancement capabilities and reconstruction procedures. Data receives classifications in multiple domains according to its purpose, where numbers or animals, or signboards represent three examples of categorized data. The enhancement technique functions to both enhance the image quality and remove unwanted noise, and it operates either on 1-D data alone or on 2-D data based on user needs. The image reconstruction process requires both the input and output neurons of the neural network to have the same number for accurate image reconstruction. The technique supports application to signals of 1-D, 2-D, and 3-D dimensions that generate outcome vectors or matrices of identical sizes to the input data. The research implements an ANN to restore visual information from the source data.The document follows a standard organization with chapters starting from the introduction to methods through results before concluding. The first part of the paper delivers a technique summary along with research reviews related to multiple uses of neural networks. The proposed approach receives a detailed explanation throughout the methods part, followed by a result presentation of images before and after the neural network technique usage. The final chapter brings together the conclusions based on the released results, which point toward prospective research agendas.

LAGOPHTHALMOS DRIVERS IN LOCOMOTIVES

Adnan Alam Khan — 2025-05-08

Every individual desires a restful night’s sleep on a comfortable bed, free from stressors. Many people invest thousands of dollars in their quest for a good night's sleep. Medical studies indicate that nearly 20% of individuals sleep with their eyes open, leading others to mistakenly perceive them as awake. This condition, known as Lagophthalmos, arises from issues with facial nerves or muscles. Symptoms include weight loss, clumsiness, eye-open sleep, and fatigue. In Western countries, those affected often adopt dogs, whose primary role is to monitor their owners. If the patient’s eyes remain open, the dog will bark loudly until they awaken. This study proposes a solution in which a camera monitors the patient via their laptop; when the patient enters an unconscious state, the system will initiate a call to their mobile phone to rouse them from sleep.

PETROGRAPHY AND GEOCHEMICAL CHARACTERIZATIONS OF TANAWAL FORMATION (QUARTZOSE ROCKS FROM SWABI, KHYBER PAKHTUNKHWA, PAKISTAN; IMPLICATIONS FOR INDUSTRIAL SUITABILITY

Muhammad Naveed Anjum — 2025-05-09

The quartzose rocks of the Tanawal Formation, exposed in Topi area, district Swabi, in Khyber Pakhtunkhwa, Pakistan. This study evaluates the studied rocks for their suitability in glass and other industries using the techniques of petrography and geochemistry. Field observations and petrographic data leads to distinction of Tanawal Formation into two discrete lithologies a) Blasto-psammites and b) Blasto-pelites. These rocks have been slightly metamorphosed and foliation planes are well developed especially in pelitic rocks. In blasto-psammites quartz, feldspars, micas, ore minerals with minor amount of tourmalines and zircon occur as the framework minerals, while in pelitic rocks quartz and micas occurs as framework minerals. Blasto-psammites have variable modal percentage of quartz (93－94%) but most of the samples fall in quartz arenite (quartzite) and sub-arkoses field in Quartz－Feldspar－Lithics (QFL) ternary diagram. Petrographic data revealed that all of the studied samples are very fine to medium grained quartzose rocks. A very low grade metamorphic event has slightly obliterated the original sedimentary fabric of these rocks but most of these are still intact. Quartz grains are present in monocrystalline as well as polycrystalline form. Geochemical analysis of the Tanawal Quartzite indicates high contents of SiO 2 (93－94 %) and fairly low percentage of other major oxide i.e MgO (0.023 wt.%), Na 2 O (2.92 wt.%), K 2 O (0.52 wt. %) thus endorsing their quartz--rich mineralogy.Both petrographic and geochemical data reveal that the studied rocks lie within the range of standard specifications required for glass industries (i.e SiO 2 93－94%). The rocks are declared suitable for the manufacturing of low quality glass products i.e. soda-lime glass, flint glass, amber glass, green glass and glass-fiber insulation

USING NLP AND AI TO ENHANCE SOFTWARE DOCUMENTATION AND CODE COMPREHENSION

Abdulmalik Ibrahim — 2025-05-10

Software documentation plays a critical role in code comprehension, maintenance, and collaboration, yet it is often incomplete, outdated, or inconsistently written. This study explores the application of Artificial Intelligence (AI) and Natural Language Processing (NLP) techniques to automatically generate accurate and context-aware documentation for software code. Leveraging transformer-based models such as CodeT5, GraphCodeBERT, and GPT-3, we developed and evaluated a system capable of producing meaningful summaries of code functions and classes. A comparative analysis between human-written and AI-generated documentation was conducted using both quantitative metrics (BLEU, ROUGE, F1) and qualitative feedback from professional developers. The results indicate that AI-generated documentation significantly improves code readability and developer efficiency, reducing comprehension time and enhancing accuracy in understanding complex code. Additionally, real-time integration of the tool within development environments proved beneficial for continuous documentation support. While AI still faces challenges in handling domain-specific code and interpreting poorly written segments, the overall impact on documentation quality is substantial. This research underscores the potential of NLP-driven tools to automate and standardize documentation practices, offering a scalable solution to one of software engineering’s longstanding challenges. Future work aims to integrate context-awareness, multilingual support, and interactive querying features to further enhance developer experience.

PREDICTING ACADEMIC SUCCESS: A MACHINE LEARNING APPROACH USING DECISION TABLES AND RANDOM FORESTS ALGORITHMS

Malak Roman — 2025-05-10

Machine Learning (ML) in educational data prediction refers to the use of AI-driven algorithms to analyses academic data (e.g., grades, attendance, engagement) and forecast student performance, identify at-risk learners, and recommend interventions. By processing historical and real-time data, machine learning (ML) models uncover hidden patterns that enable educators to optimize their teaching strategies and enhance learning outcomes. This research cones with data collected from ‘UCI Machine Learning Repository’ and the database has total of 33 attributes along 395 rows. The two classification classifiers used in this paper were Decision Table (DT") and Random Forest (RF). The best first search algorithm has been used as a preprocessing step with both classifier models. The distribution of these models is based on the analysis of the Mean Square Root Error between the predicted and actual values. The proposed decision table yields a better result as compared to the random forest algorithm with the blast 1.92 root mean squared error.

A STRIDE BASED APPROACH TO FORTIFY DIGITAL HEALTHCARE SECURITY

Summia Aziz — 2025-05-10

The study underscores the paramount significance of robust threat identification in Smart Health Systems (SHS) to enhance reliability and practicality. This study employs a methodology that effectively utilizes the Microsoft Threat Modelling Tool, leveraging the STRIDE framework, to comprehensively identify threats within the SHS. Through the incorporation of the DREAD model, this approach facilitates efficient risk management, prioritizing threats based on Damage, Reproducibility, Exploitability, Affected Users, and Discoverability. The systematic framework proposed in this research proves instrumental in identifying threats at early stages, streamlining the mitigation process, and contributing to the overall security resilience of Smart Health Systems. This work aims to fortify the foundation of SHS by addressing security challenges proactively, ensuring a reliable and secure digital healthcare ecosystem.

EDGE INTELLIGENCE IN IOT: ENABLING SMARTER, FASTER AUTONOMOUS DEVICES THROUGH ARTIFICIAL INTELLIGENCE AND EDGE COMPUTING

Idrees Mustafa — 2025-05-12

The intersection of Edge Computing and Artificial Intelligence represented by Edge Intelligence (EI) is paving a new road for the Internet of Things (IoT), eliminating its main shortcoming, i.e. the latency, the bandwidth constraints, and the privacy concerns of the traditional cloud systems. In this paper, we have discussed principles, challenges and applications of EI, focusing on its overall contribution to real-time secure and energy-efficient operations in healthcare, smart cities, industrial IOT and autonomous vehicles. Federated learning, lightweight AI models, and hybrid edge-cloud architectures are analyzed on the grounds of their key technological advancement on how energy and scalability restrictions can be overcome. Finally, integration of 6G networks with blockchain technology along with an ethical AI framework is proposed as a path to enable future capabilities. This work intends to point researchers, developers, and policymakers in the direction of adopting security and sustainability in EI system by providing a comprehensive survey of the available solutions and future trends.

DESIGN AND PERFORMANCE ANALYSIS OF A MULTILEVEL INVERTER USING A BIPOLAR PWM SWITCHING SCHEME

Muhammad Essa Majeed — 2025-05-12

This paper focuses on the design and simulation of a single-phase PWM multi-level inverter using the bipolar switching scheme to improve efficiency and quality with the help of pulse width modulation (PWM) for DC to AC voltage converters. This research emphasizes on obtaining well defined output voltage, reduction of total harmonic distortion (THD), and enhancing energy conversion effectively and efficiently by utilizing multilevel inverter and phase disposition (PD) modulation techniques. The inverter shows outstanding results, achieving an efficiency of 96% and producing an output voltage of 110-Volts at 60 Hz from the 80-Volt DC input from each bridge. In this particular simulation in MATLAB, the inverter demonstrates a significant reduction in THD, with an output voltage THD of 1.36% and a current THD of 0.13%. Our approach is effective and has the potential to produce stable AC output even with RL loads and maintain high power quality. The paper highlights the significance of modulation techniques in increasing the performance of inverters and also underscores the importance of selecting suitable methods for different applications to ensure the best performance and efficiency.

NEXT-LEVEL SYSTEM DESIGN: ADVANCED AND HIGH-PERFORMANCE SYSTEM ARCHITECTURES FOR THE FUTURE OF ELECTRIC VEHICLES.

Sheeraz Ahmed — 2025-05-13

The transformation of the automotive industry toward electric mobility necessitates the development of advanced and high-performance system architectures to address the evolving demands of next-generation electric vehicles (EVs). With an increasing focus on sustainability, efficiency, and advanced technological integration, future EV platforms must support cutting-edge capabilities such as autonomous driving, smart energy management, and seamless connectivity. This paper investigates the design and integration of next-level system architectures that will define the future of electric vehicles. Emphasizing the need for modular, scalable, and fault-tolerant systems, the research explores how these architectures can facilitate continuous innovation, reduce vehicle obsolescence, and improve safety and reliability. Key topics include the integration of high-efficiency powertrains, advanced battery technologies, and intelligent control systems that enhance vehicle performance and user experience. Furthermore, the study highlights the role of emerging technologies such as artificial intelligence, machine learning, and vehicle-to-grid (V2G) communication in optimizing system performance, vehicle autonomy, and energy efficiency. The paper also examines the potential for adaptive architectures that can accommodate new functionalities and technological upgrades over the vehicle's lifecycle. By exploring these innovations, this research provides insights into how next-generation system designs will help electric vehicles meet the growing global demands for performance, safety, and environmental sustainability. The findings aim to guide the development of robust, future-proof EV architectures that will shape the automotive industry’s transition to a more sustainable and technologically advanced future.

INVESTIGATING THE INFLUENCE OF CONTINUOUS INTEGRATION ON SOFTWARE QUALITY AND DEVELOPER PRODUCTIVITY

Faryal Gul — 2025-05-13

Continuous Integration (CI) has emerged as a pivotal practice in modern software development, impacting both software quality and developer productivity. This study conducts a systematic literature review (SLR) to investigate the reported claims regarding the effects of CI on software development processes. The synthesis of findings from diverse sources provides a nuanced understanding of CI practices, tools, and their influence on software quality parameters and developer productivity. The SLR focuses on key aspects, including code stability, bug detection, release confidence, collaboration, issue resolution, and documentation. The study uncovers insights into the multifaceted role of CI in shaping software quality and explores its implications for developers working in various environments. Additionally, the research identifies challenges, contributions, and limitations within the existing literature. While the study contributes valuable insights, it recognizes certain limitations, such as the dynamic nature of CI practices and the heterogeneity of development environments. The findings highlight the need for continuous monitoring of emerging trends, empirical validation of reported claims, and exploration of the integration of CI with emerging technologies. This study provides a comprehensive overview of the influence of CI on software development, contributing to the ongoing discourse on effective software engineering practices. The identified challenges and avenues for future research guide the way for further exploration, refinement, and adaptation of CI practices in the ever-evolving landscape of modern software development.

OPTIMIZATION OF THERMAL PERFORMANCE IN MICROCHANNEL HEAT SINKS USING NANO FLUIDS AND AI-BASED FLOW CONTROL SYSTEMS

Muhammad Aqeel — 2025-05-13

Introduction The optimization of thermal performance of microchannel heat sinks is important to improve the efficiency of cooling systems in electronics and industrial cooling applications. Nanofluids, i.e. fluids If you have windows XP) Fluids that have been enhanced with nanoparticles have achieved much research interest because of their enhanced heat transfer capabilities. Furthermore, employing AI-based flow control systems, like reinforcement learning, could introduce a new degree of freedom for online optimization of heat sink systems. Objectives: The main goal of this work is to examine the thermal performance of the microchannel heat sinks for the applications of the nanofluids (Al2O3, TiO2 and CuO) under various volume fractions optimized with AI based flow control to improve the heat transfer efficiency with the reduced rate of energy utilization and pressure drop. Method: The microchannel heat sink experiment and were used to experimentally characterize the nanofluids with various volume fractions. The flow rates and pressure drops were measured and the thermal performance was analyzed based on Nusselt number, thermal resistance and heat transfer coefficient. Reinforcement learning algorithms for autonomous control based on AI were used for the real-time control on the flow rate. Computational fluid dynamics (CFD) simulations were also performed to compare with experiment. Results: It was found that nanofluids enhanced the heat transfer whether were more effective than base fluids, Al₂O₃ having gave best performance. The thermal efficiency was optimized by AI-driven flow control, which decreased the temperature by 12.3 °C and the pressure drop by 110 Pa compared with fixed flow rate systems. Nusselt numbers were the CFD simulations performed with good accuracy in the predictions having error of using 2.1%. Conclusions: The joint use of nanofluids and AI-based flow controls systems can be considered an efficient and compact technology for improving the thermal performance of microchannel heat sink, thus making it a promising cooling solution for the new century's applications.

DESIGN AND IMPLEMENTATION OF AN AI-DRIVEN OBJECT TRACKING ROBOT WITH HUSKY LENS INTEGRATION

Afnan Ahmed — 2025-05-14

The era of Artificial Intelligence has arrived, with robotics as its most significant application. This paper introduces an object-tracking AI robot programmed to follow a specific object. The robot uses the AI-powered camera HuskyLens to sense and track the objects. The robot is such that it is first trained to learn the object. Once discovered, the robot is able to track and follow the specific object. The HuskyLens senses the object, and the information is passed to the Arduino, which then enables the motors via the motor driver to follow the object. Some existing techniques for controlling object tracking robots are IR sensors, LDR, and RGB sensors. As most of these sensors are not designed for object tracking, these techniques are often not useful enough to meet the requirements of accurately tracking the object. Huskylens, with its AI-powered smart camera alone, can be used as the only sensor in the system to control movement and fulfill the functionality required for object-tracking robots and their various applications.

PATTERN IDENTIFICATION OF DRUG RESISTANCE FOR TUBERCULOSIS IN PAKISTAN USING MACHINE LEARNING TECHNIQUES

Omaid Ghayyur — 2025-05-14

Tuberculosis (TB) remains a global health challenge due to the rise of drug-resistant strains, particularly multidrug-resistant TB (MDR-TB). This study employs machine learning to predict drug resistance patterns in TB patients using clinical data from Pakistan. We collected a dataset of 400 pre-processed samples with 12 key features, including demographic and drug response data, from multiple regions in Pakistan. After preprocessing and addressing class imbalance using the Adaptive Synthetic Sampling (ADASYN) technique, we evaluated nine supervised learning algorithms Multi-Layer Perceptron, Decision Tree, Random Forest, Naïve Bayes, Support Vector Machine, Gradient Boosting, Extreme Gradient Boosting, Logistic Regression, and an ensemble model under three techniques: Whole Dataset Imbalanced (Technique 1), Training Dataset Balanced with ADASYN (Technique 2), and Whole Dataset Balanced with ADASYN (Technique 3). Results show that NB achieved the highest realistic accuracy of 96.55% under Technique 2, with DT, RF, and the Ensemble model at 94.83%. Under Technique 3, NB reached a peak accuracy of 99.61%, outperforming prior literature benchmarks. These findings highlight the competitive performance of machine learning in the early detection of TB drug resistance, offering a pathway to improve treatment outcomes in resource-limited settings.

EDGE INTELLIGENCE IN IOT: ENABLING SMARTER, FASTER AUTONOMOUS DEVICES THROUGH ARTIFICIAL INTELLIGENCE AND EDGE COMPUTING

Idrees Mustafa — 2025-05-14

REVOLUTIONIZING TELECOMMUNICATIONS: THE IMPACT OF IOT

Abdullah Haseeb — 2025-05-14

The internet of things (IoT) has emerged as a transformative pressure in the telecommunications zone, revolutionizing how networks are based, operated, and utilized. As the number of connected devices continues to grow exponentially, telecommunication networks must evolve to satisfy the needs of this hyper-connected environment. This observation explores the profound effect of IoT at the telecommunications enterprise, specializing in the way it drives innovation, optimizes community performance, and allows new enterprise models. The IoT is predicated closely on sturdy telecommunications infrastructure to make certain seamless verbal exchange between billions of devices. The arrival of the 5G era plays an important role in this modification, presenting ultralow latency, excessive bandwidth, and the potential to support massive tool connectivity. As a result, telecom operators are moving from conventional communique models to IoT-specific services, including low-energy extensive-vicinity networks (LPWAN) and better facts analytics talents. Furthermore, IoT programs, ranging from smart homes and cities to business automation and healthcare, place a giant strain on telecom networks to deliver faster, more reliable, and greater relaxed connectivity. Telecommunications groups are responding by investing in network virtualization, side computing, and records analytics, which permit them to address the surge in IoT visitors and offer real-time insights for companies and purchasers. The examine additionally examines the emerging opportunities for telecom providers, inclusive of new sales streams via IoT as a provider (IoTaaS) and partnerships with IoT device manufacturers. Moreover, it addresses the challenges of scalability, safety, and information privacy in IoT ecosystems, emphasizing the role of telecom in ensuring the strong safety of interconnected gadgets. In the end, the IoT is essentially reshaping telecommunications, supplying enormous demanding situations and thrilling opportunities for growth. Collaboration among telecom operators, IoT developers, and clients will ultimately decide the achievement of this technological revolution.

A COMPARATIVE STUDY OF BLOCKCHAIN-BASED E-VOTING SYSTEMS: CHALLENGES, TECHNIQUES, AND IMPLEMENTATION

Syed Shabeeb Raza — 2025-05-14

Traditional voting methods require the physical appearance of voters. With the progress of digital technology and the advent of electronic voting systems, voters can now vote from remote locations. However, even such systems have to face many challenges in terms of safety and privacy. Here, we propose our e-voting system based on blockchain technologies to ensure voter information's anonymity, security, and consistency through Merkle trees and hash digests. The data alteration is easily detected, leading to a compromised block's rejection. This research introduces a novel e-voting solution using the innovative approach of blockchain technology, applying the Advanced Encryption Standard (AES) and Zero Knowledge proof algorithms. Our decentralized architecture ensures a highly secure and transparent voting process and provides a robust framework for verifiable and auditable elections. Applying advanced cryptographic techniques guarantees the confidentiality and integrity of each vote. Designed to be user-friendly, accessible, and adaptable for organizations with internal polls. This landmark initiative lights the way to a new voting process, making it different and impossible to counter until now, as it focuses on security, transparency, and accessibility- all of these being a giant leap forward in the evolution of electronic voting technology.

INTRUSION DETECTION IN IOT THROUGH DEEP LEARNING

Muhammad Awais — 2025-05-15

Intrusion detection forms an important element to security in the realm of Internet of Things since it aids early identification of any incidence to compromise with the system. In the case of IoT contexts, intrusion detection processes are often futile due to the sheer volume of data produced by these gadgets. Hence, deep learning and several state-of-the-art and effective intrusion detection technologies that is required. The deep learning domain of machine learning applies neural networks to a huge amount of information. It has been proven exceptionally effective in various applications such as, audio and picture recognition, natural language processing and anomaly detection. In terms of IoT security, deep learning can be employed to detect odd behavior in a device’s activity and enable timely detection of potential security abnormalities. Thus, this research has great importance as it can help to minimize dangers associated with IoT gadgets and protect confidentiality and personal information of users. Deep learning algorithms are expected to be the solution to the intrusion detection problem in the Internet of Things. In fact, a recurrent neural network (RNN) is used to recognize abnormal behavior by first training on the normal activity of large numbers of devices. As this model has been trained using an unsupervised learning method (i.e.,without labelled data), it is particularly suited to the challenges of security in the world of Internet-of-Things technology. Implementing this solution involves collecting and cleaning data from Internet of Things devices, then training the RNN model to detect abnormalities. The model is evaluated using standard measures such as accuracy, recall and F1 score. Analyzing the results, it seems that deep learning techniques can indeed be used to identify aberrant behavior among Internet of Things (IoT) devices. It would seem then that they could play a role in hardening IoT device security and reducing attack risk.

TECHNICAL EVALUATION OF THE DARWAZA FORMATION IN ENGINEERING AND CHEMICAL INDUSTRIES, KHYBER PAKHTUNKHWA, PAKISTAN: INSIGHTS FROM PETROGRAPHY, GEOTECHNICAL, AND GEOCHEMICAL PROPERTIES

Muhammad Naveed Anjum — 2025-05-15

The objective of this research is to conduct a comprehensive geotechnical investigation of the Darwaza Formation at the study site, a Cambrian-aged geological unit predominantly composed of limestone. Petrography and a series of physio-mechanical, chemical, and analytical tests were conducted, including Los Angeles (LA) abrasion, impact strength, specific gravity, water absorption, soundness testing, scanning electron microscopy (SEM) with attached energy dispersive X-ray spectroscopy (EDX). Petrographic analysis of the Darwaza Limestone unit under polarized light microscopy reveals that the limestone is largely micritic to sparitic in texture, with varying degrees of recrystallization. The primary constituents include calcite, with minor occurrences of dolomite in some samples. The calcite in the Darwaza Limestone exhibits neomorphic textures, indicating diagenetic alteration of the original micritic material. Mineralogical characteristics and textural features were correlated with physio-mechanical properties to validate the laboratory results. The physio-mechanical properties (e.g., LA =20.9%, Soundness = 2.0%) and Impact value test = 12.1%), of the Darwaza Formation are found to fall within acceptable standard limits of ASTM. Based on these findings, it is proposed that the limestone of the Darwaza Formation is suitable for engineering applications and can be effectively used as construction aggregate. Beyond its suitability for the construction industry, the high calcium carbonate (CaCO₃) content—exceeding 95% also supports the potential use in paper, glass, sugar, ceramics, adhesives, food and pharmaceutical industries.

ENHANCING AI SYSTEM TRANSPARENCY AND EXPLAINABILITY: INTEGRATING FORMAL METHODOLOGIES FOR IMPROVED MODEL PERFORMANCE AND INTERPRETABILITY

Sultan Salah Ud Din — 2025-05-15

Artificial Intelligence operates as essential business infrastructure in healthcare together with finance and autonomous systems. The output decisions from deep learning neural network-based AI models present significant barriers to both understanding and interpretation. The absence of explainability features between models creates trust-related conflicts for users and regulators and directly affected industrial stakeholders. The combination of SHAP and LIME presents viable explanation tools but produces imprecise interpretations when evaluated against high-dimensional real-time datasets. Random Forest surpassed both Logistic Regression and SVM by obtaining superior results in generalization testing which produced greater training and validation accuracy levels. The accuracy measurements revealed that Random Forest achieved 0.894 training accuracy along with 0.879 validation accuracy while Logistic Regression maintained 0.905 training accuracy and 0.874 validation accuracy and SVM achieved 0.848 training accuracy with 0.867 validation accuracy. The decision outcomes from the model were primarily influenced by Features 3 and 6 according to SHAP and LIME analysis. Random Forest presented the best ROC and precision-recall curves which indicated its strength to separate distinct classes. Future research should optimize the methodologies through development that enables their scaling across multiple applications while achieving better performance specifically in time-sensitive and dimensionally complex systems. Despite these promising results, the study encountered two primary limitations: Formal methods face scalability issues and all models displayed poor AUC scores as their primary limitations. Both Logistic Regression and Random Forest with SVM yielded prediction performance similar to random guessing based on AUC scores of 0.51 and 0.50 respectively. The research focus should optimize scalable methods aimed at improving performance while solving time-sensitive high-dimension problems.

INVESTIGATING THE IMPACT OF MATERIAL FATIGUE ON STRUCTURAL IN HIGH-PERFORMANCE MECHANICAL SYSTEMS

Muhammad Aqeel — 2025-05-15

Background: The structural integrity and operational reliability of high-performance mechanical systems are substantially impacted by material fatigue. This knowledge is vital for improving and prolonging the performance and lifespan of systems that face fatigue failures, especially as demands for a higher level of safety and durability are being placed on many engineering applications. Severe fatigue failure prediction and prevention have been extensively studied, yet accurately predicting fatigue life and establishing effective prevention methods remain challenging. Objective: This study aims to analyze the primary causes of material fatigue, evaluate existing mitigation techniques, and explore the potential of emerging technologies in enhancing fatigue resistance. Methods: A systematic survey was conducted among 135 professionals in the mechanical systems engineering field, including engineers, scientists, researchers, and technicians with diverse expertise levels. The survey investigated industry trends, key contributions, and perspectives on material fatigue mitigation. Additionally, a literature review was performed to analyze recent advances in material science and fatigue prediction models to contextualize the findings within the broader research landscape. Key Findings:

Repetitive loading conditions were identified as the primary contributor to material fatigue, as reported by 75% of respondents.
Environmental effects were another significant factor, with 60% of respondents acknowledging their impact on material degradation.
More than two-thirds (67%) of respondents had encountered structural failures due to fatigue in their professional experience.
Preventive strategies such as regular inspections and high-quality materials were found to have only moderate effectiveness.
Smart monitoring systems and computational fatigue models showed great potential but faced challenges related to technological constraints and cost. Conclusion: To address material fatigue challenges in mechanical systems, improved predictive maintenance strategies and the development of novel materials are essential. Future research should focus on developing fatigue-resistant alloys, self-healing materials, and AI-driven monitoring systems to enhance structural durability. Computational modeling and real-time data analytics will play a critical role in understanding fatigue progression and defining future engineering solutions, ultimately elevating industry standards.

SMART ATTENDANCE MANAGEMENT SYSTEM WITH FACIAL RECOGNITION, PHYSICAL PRESENCE VERIFICATION, AND CENTRALIZED DATABASE INTEGRATION

Awais Gul Airij — 2025-05-15

The developed Smart Attendance Management System, hereinafter referred to as ‘the System’, addresses a limitation faced in Facial Recognition Systems (FRS), which often fails to authenticate physical presence during attendance marking. The system attempts to solve this problem by designing a solution that authenticates physical presence while attendance is being marked. The solution uses physiological signals, specifically the blinking of eyes, to indicate physical presence. The system marks attendance by first matching the images stored in its database with input images. Upon successful matches, the system looks for physical presence authentication. Attendance is marked successfully upon verification of physical presence. The attendance is recorded in a database for administrative reference. The designed system uses visual representations to interact with users during the entire process of attendance marking for both successful and unsuccessful marks. The system offers reliable attendance tracking by combining eye blinking detection with face recognition algorithms to provide a foolproof attendance management system.

SIMULATION-BASED STUDY OF VOID-CONTROLLED PRECAST SQUARE CONCRETE-FILLED STEEL TUBE COLUMNS UNDER AXIAL COMPRESSION

Jamshad Alam — 2025-05-16

This research explores the axial compressive performance of void-controlled precast square CFST columns via experimental testing and finite element analysis (FEA). Conventional CFST columns are plagued by interfacial defects and voids, undermining structural performance. This work suggests a new concept by introducing controlled voids into precast square CFST columns with high-performance self-compacting grout to improve load-carrying capacity and material efficiency. A confirmed finite element model, designed by applying ABAQUS/Standard 2024, perfectly mimics nonlinear interaction among steel confinement, concrete crushing, and grout expansion, with the outcomes reflecting less than 5% deviation from experiments. Parametric studies demonstrate that a 20 mm reserved gap filled with C40 grout enhances ultimate bearing capacity by 4.55% compared to solid columns and lessens structural weight up to 30%. The research shows that up to 20% void ratios maintain more than 95% of the solid column capacity, with grout expansion (0.02–0.05% strain) reducing interfacial debonding. A new load-capacity equation that considers void reduction factors and grout confinement effects is introduced, providing an easy-to-use design tool for engineers. The research shows the possibility of void-controlled precast CFST columns as an efficient and environmentally friendly solution for future construction that weighs performance, constructability, and cost savings. Cyclic/seismic performance and long-term durability are potential future directions of research.

EXPLORING THE POTENTIAL OF BLOCK CHAIN AND AI CONVERGENCE TO SECURE AND VERIFY IOT DATA TRANSMISSIONS IN HIGH-STAKES INDUSTRIES LIKE HEALTHCARE AND FINANCE

Muhammad Ismail — 2025-05-16

This paper evaluates Block chain technology and Generative AI solutions for healthcare alongside their healthcare implementation challenges and solution methods. Block chain technology strengthens electronic health record security alongside privacy protection and stands as a solution for improving record interconnection. Smart contracts simultaneously modernize supply chain process management and administrative functions. Block chain technology proves able to authenticate IoT data while protecting its security thus delivering improved medical services through documented case examples. The healthcare industry experienced major transformation through ChatGPT Generative AI which now enables doctors to deliver personalized treatment along with diagnostic tests and predictive outcomes. AI systems process large datasets which helps medical staff diagnose diseases earlier while simultaneously generating treatments personalized for each patient. The implementation of AI-based virtual healthcare assistants has raised patient involvement through real-time information services and better treatment following procedures coupled with continuous user assistance. Generative AI has accelerated medical research along with drug development processes which shortens the duration and decreases expenses involved in medication production. Implementations of Generative AI alongside Block chain systems enable protected patient data storage methods as well as advanced AI training resources and streamlined health operation methods. The existing challenges revolve around scalability and energy consumption together with interoperability issues. The paper creates plans for expandable block chain systems along with controlled information exchange protocols. Medical research together with treatment benefits from these technologies since they improve both safety measures and targeted treatment approaches.

AI IN ARCHITECTURE AND ACADEMIA: BRIDGING INNOVATION AND ETHICS FOR SUSTAINABLE HUMAN-CENTRIC DESIGN

Ar. Uffaq Shahid — 2025-05-16

Artificial Intelligence (AI) is reshaping academia and architectural practice by augmenting creativity, optimizing workflows, and fostering interdisciplinary innovation. While AI-driven tools enhance design processes, research methodologies, and educational paradigms, ethical concerns, workforce adaptation, and data governance challenges persist. This paper explores AI’s dual role in these domains, evaluates its implications, and proposes frameworks for sustainable integration.

Drawn objects have no meaning in the virtual realm of 3D visualization software, and they are the short output of binary data on a screen. Users, on the other hand, perceive such items as buildings, rooms, architectural elements, and so on. One of the problems impeding the integration of digital and sustainable design is the recognition gap between computers and humans. Obstacles include accurate realistic iterations of massive calculations. Accuracy and practicality, including spatiality, materiality, and so on, are critical considerations in environmental simulations. Without these factors, it is impossible to replicate environmental performance in any situation.

To obtain relevant results, a large amount of data must be determined for computations in environmental modelling software such as ECOTECT, TAS, and others. User choice is another impediment to not just environmental optimization but also architectural design concerns. It may be simple to meet objectives just on a numerical basis.

For example, getting the most sunshine is as simple as using the largest window or glass box. These options, however, are, needless to say, unacceptable to designers or architects. Computer-generated environmental solutions are only beneficial if the users are happy with them. Furthermore, the presence of clients complicates initiatives. As a result, optimum solutions must be created in line with user choices.

The intersection of AI with academia and practice highlights a dual-edged impact—while it offers immense potential for growth and efficiency, it demands careful navigation of ethical dilemmas and societal implications. Institutions must foster interdisciplinary collaboration to address these challenges while maximizing the benefits of AI. As AI continues to evolve, its role in shaping the future of education and professional practice will be pivotal in driving innovation while ensuring inclusivity and ethical responsibility. this Abstract Underscores the transformative potential of AI while emphasizing the importance of addressing its broader implications for sustainable development in both academic and practical domains.

XCEPFOREST: A HYBRID DEEP LEARNING AND MACHINE LEARNING APPROACH FOR ENDANGERED AQUATIC SPECIES CLASSIFICATION

Khalil Haider — 2025-05-16

From the appearance of Homo sapiens until now, innumerable species have gone extinct, and many marine organisms are now on the verge of extinction due to human activity and changing environmental conditions. In this study, we proposed a hybrid model for endangered aquatic species classification based on deep learning and machine learning. We use the Xception Pre Trained Model for high-level image feature extraction, followed by Random Forest classification and termed it “XcepForest”. We utilized the data downloaded from iNaturalist which was annotated with video frames extracted from the trusted accounts, and performed bilateral filtering and adaptive gamma correction for better image quality before feeding into the model. Across an 80-20 train-test split, XcepForest achieved 92.73% accuracy on the training set and 92.45% accuracy on the test set, successfully annotating species including the green sea turtle, whale shark, and blanket octopus. Overall, the majority of existing research has focused on the classification of fish and other marine mammals, and often requires high computational power and does not aim for the monitoring of endangered species. Thus combining deep feature extraction with a machine learning classifier, provides an efficient and scalable solution to the conservation of marine biodiversity.

FORTIFYING BITCOIN TRANSACTIONS: ADVANCED MACHINE LEARNING TECHNIQUES FOR FRAUD DETECTION

Muhammad Haris Masud — 2025-05-17

With the growing adoption of cryptocurrencies, Bitcoin has emerged as a prominent player in the global financial landscape. However, its decentralized and pseudonymous nature has made it an attractive target for fraudulent activities. This paper presents a comprehensive exploration of fraud detection techniques specifically tailored to Bitcoin transactions.

In this research, we delve into the intricacies of the Bitcoin network, analyzing transaction data, and identifying patterns that indicate potential fraudulent behaviour. We propose a multifaceted approach that combines machine learning algorithms, graph analysis, and heuristic rule-based systems to detect various types of fraud, including Ponzi schemes, money laundering, and unauthorized transfers.

Our study leverages the transparency of the blockchain to extract relevant features and build models capable of identifying anomalous transactions. Furthermore, we address the challenges posed by the dynamic nature of Bitcoin transactions, such as mixing services and privacy enhancements, which attempt to obfuscate transaction trails. We discuss strategies for adapting our fraud detection techniques to these evolving tactics, ensuring the continued effectiveness of our approach.

To validate our methodology, we present empirical results based on a comprehensive dataset of real-world Bitcoin transactions. We demonstrate the efficacy of our approach in detecting fraudulent activities and showcase its potential to enhance the security and trustworthiness of Bitcoin as a digital asset.

In conclusion, this paper contributes to the growing body of research aimed at safeguarding the integrity of cryptocurrency networks. By proposing advanced fraud detection techniques tailored to Bitcoin transactions, we take a significant step toward mitigating the risks associated with cryptocurrency use, fostering trust among users, and facilitating its broader adoption in the global financial ecosystem.

ANALYSIS OF AIRFOIL USING WIND TUNNEL THROUGH IMAGE PROCESSING

Abdul Sami — 2025-05-17

This project introduces a novel approach to measuring aerodynamic parameters using digital image processing, offering a simpler and more cost-effective alternative to traditional wind tunnel methods involving manometers and anemometers. The study explores three main methodologies, including Computational Fluid Dynamics (CFD), pressure sensors, and image processing, with a focus on the latter. By recording airflow behavior over an airfoil using a high-quality digital camera and analyzing the footage in MATLAB, pressure and velocity were determined effectively. A smoke generator was used to visualize airflow patterns in the wind tunnel. The image processing results showed strong agreement with anemometer readings and CFD simulations, confirming its reliability. This method simplifies aerodynamic testing and offers a more efficient alternative to conventional techniques.

PARAMETRIC STUDY ON THE ROLE OF SUPPLEMENTARY MATERIALS IN MODULATING THE PERMEABILITY COEFFICIENT OF GEOPOLYMER CONCRETE

Amjad Ali — 2025-05-19

Geopolymer concrete represents an emerging trend within the concrete industry, serving as a potential alternative to traditional cement concrete. It offers numerous advantages compared to conventional cement-based materials. Geopolymer concrete is an environmentally friendly concrete composed of waste materials produced by thermal plants in various power generation industries. The alkaline activators in geopolymer concrete exhibit excellent heat resistance and strong adhesion between aggregates, leading to superior mechanical properties. Furthermore, geopolymer concrete exhibits significantly greater durability compared to conventional cement concrete. This study utilized marble waste and fly ash as raw materials for the production of geopolymer concrete. The materials were subjected to identical curing conditions to assess their mechanical properties. The investigation's results indicated that geopolymer concrete based on Marble can be produced under various curing conditions, demonstrating excellent physical and mechanical properties. The tests to evaluate the mechanical properties included split tensile, compressive, and water penetration assessments. The optimum mixes utilized in this research include 100% fly ash, 75% fly ash and 25% marble powder, 50% fly ash and 50% marble powder, and 25% fly ash and 75% marble powder. The alkaline activator to binder ratio is 0.40 [7]. The research focused on investigating geopolymer concrete's mechanical and physical properties.

SECURING AI-DRIVEN TEXT CLASSIFICATION AGAINST ADVERSARIAL NLP ATTACKS

Komal Azim — 2025-05-19

The integration of Artificial Intelligence (AI) has revolutionized Natural Language Processing (NLP) enables advanced text classification tasks such as sentiment analysis, spam detection, and news categorization. However, the widespread adoption of AI in NLP has introduced significant cybersecurity risks, as these systems are highly vulnerable to adversarial attacks. These attacks aim to skew predictions and compromise their accuracy and integrity by making minor adjustments to input data, taking advantage of flaws in NLP models. We analyse and assess adversarial assaults on text categorization methods using AG News datasets. We examine how the model's performance might be assessed without human visual notice by implementing relatively straightforward transformation techniques such word substitution, paraphrase, or syntax alterations. These attacks highlight the basic flaws in NLP systems and demonstrate how easily they may be twisted and used maliciously. With up to 97% resilience against hostile attacks, the models proposed ensemble models by integrating the deep learning architectures include Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM), and Recurrent Neural Networks (RNN). CNN performed better at identifying localized features, even though both the LSTM and RNN models showed good sequential processing skills. They significantly increased their resilience by combining complimentary qualities into ensemble frameworks. The highest success rate demonstrates that the ensemble tactics work to reduce adversary manipulation while preserving excellent classification accuracy.

PREDICTIVE MODELING OF URBAN AIR QUALITY IN KARACHI USING MACHINE LEARNING AND OPEN-SOURCE SATELLITE DATA

Danish Mustafa Khan — 2025-05-20

This research aims to develop a predictive AI model to forecast and monitor air quality in Karachi by utilizing publicly available environmental and satellite datasets, significantly eliminating the dependency on non-reliable extensive physical sensor infrastructure. The study leverages data from Copernicus Atmosphere Monitoring Service (CAMS), OpenAQ and NASA's MODIS, analyzed with meteorological inputs from the Pakistan Meteorological Department (PMD). Supervised learning techniques, including LSTM neural networks and Random Forest, are used to analyze concentrations of PM2.5, PM10, and NO₂ in relation to humidity, PH, temperature, urban activity proxies and wind patterns. The impact of seasonal events like monsoon winds, traffic surges and smog are also examined. The final goal is to deliver forecasts and real-time air quality alerts through digital platforms, significantly contributing to public health resilience and smart city development in Karachi.

VOICE-ACTIVATED SMART ENVIRONMENTS: DEEP LEARNING APPROACH FOR PASHTO SPEECH COMMAND PROCESSING

Masood Anwar — 2025-05-20

Modern Automatic Speech Recognition-ASR systems leverage deep learning architectures like transformer-based models to convert spoken language into text with human-level accuracy. The integration of a command extraction controller enables real-time parsing of semantic intent, transforming raw audio signals into executable instructions for IoT devices, robotics, or assistive technologies. This dual-stage pipeline—combining acoustic modeling with context-aware natural language understanding (NLU). This article describes the creation and deployment of a Speech Recognition and Command Extraction Controller-integrated Automatic Speech Recognition (ASR) system for the Pashto language. By utilizing NLP technology, the Speech Recognition Controller was easily included into the system, improving the comprehension of user commands. The use of appliance controls showed effective command execution and security features. The command list worked well and gave users precise instructions. Entirely 150 different participants participated in the dataset gathering process to guarantee that the system would work with a range of voices, accents, and speech patterns. With an overall performance parameter of 92%, which indicates good accuracy (94%), precision (92%), recall (90%), and F1-score (92%), the system's overall performance was reliable and effective. The system's speed, dependability, and efficiency in interpreting and comprehending Pashto instructions make it a workable option for a range of applications using Pashto voice recognition.

EVALUATING THE EFFECTIVENESS OF TRANSFER LEARNING IN FEW-SHOT LEARNING SCENARIOS FOR NATURAL LANGUAGE PROCESSING TASKS

Dr. Naeem Fatima — 2025-05-20

This study investigates the effectiveness of transfer learning in few-shot learning scenarios across various natural language processing tasks. The research systematically evaluates three pre-trained language models (BERT, RoBERTa, and T5) across five NLP tasks with limited training data. Through rigorous experimental analysis involving varying training set sizes from 10 to 100 examples, the study demonstrates that transfer learning substantially improves performance in data-scarce environments compared to models trained from scratch. Results indicate that RoBERTa consistently outperforms other models across most tasks, with performance gains becoming more pronounced as training examples increase from 10 to 100. Task-specific analysis reveals that sentiment analysis and text classification benefit more from transfer learning than complex tasks like summarization. The research also identifies a performance plateau effect where gains diminish beyond certain data thresholds, suggesting opportunities for more efficient fine-tuning strategies. These findings provide valuable insights for practitioners implementing NLP solutions under data constraints and contribute to the broader understanding of transfer learning dynamics in few-shot learning contexts.