Spectrum of Engineering Sciences

DESIGNING AN INTELLIGENT SMART VTOL DRONE USING FLYSKY TRANSMITTER AND PIXHAWK CONTROLLER FOR HIGH-RISK RESCUE AND MILITARY OPERATIONS

2025-03-15T09:01:01+02:00

The development of intelligent unmanned aerial vehicles (UAVs) has revolutionized the field of high-risk rescue and military missions, offering enhanced capabilities in remote and hazardous environments. This paper presents the design and implementation of a smart Vertical Take-Off and Landing (VTOL) drone, utilizing a FlySky Transmitter and a Pixhawk Controller for autonomous flight operations. The proposed drone integrates advanced navigation and control algorithms, enabling seamless performance in dynamic and complex terrains. The FlySky Transmitter ensures robust communication between the drone and the operator, while the Pixhawk Controller manages the flight stabilization, sensor integration, and autonomous mission execution. The system is designed to support both manual and autonomous flight modes, with intelligent features such as obstacle detection, real-time video streaming, and automatic emergency procedures. Through extensive flight testing, the drone's performance is evaluated in various high-risk scenarios, demonstrating its potential to enhance mission effectiveness and safety in critical applications such as search and rescue operations and military surveillance. The findings highlight the system's reliability, adaptability, and potential for future integration with advanced AI technologies, positioning it as a valuable tool for high-risk environments. The paper concludes with a discussion on the future prospects of smart VTOL drones, addressing both technical challenges and ethical considerations in their deployment.

A STUDY ON THE DETECTION AND PREVENTION OF CYBER ATTACKS USING MACHINE LEARNING ALGORITHMS

2025-03-15T09:13:16+02:00

This study explores the use of machine learning algorithms to detect and prevent cyberattacks. The research focuses on several widely used models, including Decision Trees, Support Vector Machines (SVM), Random Forests, and Neural Networks, evaluating their performance on datasets related to network traffic, intrusion detection, and malware classification. Preprocessing techniques such as data cleaning, feature selection, and balancing were applied to optimize the datasets for model training. The results show that Neural Networks outperformed the other algorithms in terms of accuracy, precision, recall, and F1-score, followed by Random Forests. This study highlights the importance of machine learning in cybersecurity, demonstrating its potential to detect complex attack patterns and improve real-time threat detection systems.

COMPARATIVE PERFORMANCE ANALYSIS OF TEXTURED AND UN- TEXTURED TURNING INSERTS IN DRY MACHINE OF AL7075

2025-03-15T09:21:26+02:00

Surface texturing is an effective process for improving the tribological characteristics Surface roughness may be beneficial in a number of ways, including decreased friction and tool wear due to the reduced contact area and formation of micro pools of lubricant in textures.

In machining operations the tool shape material and cutting parameters must all be carefully chosen. Thesevariables can influence both the qualityof work piece and the tool wear during machining. Simple and textured (Parallel and perpendicular grooved ,dimple textured) Tungsten Carbide inserts were utilized for machining of Al-7075 in this study. L20 (5*4) Taguchi orthogonal array was used for experimental design. CMM and Surftonic surface meter was used to measure tool wear and surfaceroughness respectively. The efficiency of this strategy is demonstrated by the Pareto coefficient main effect plot, contour plot, and surface plots. The study found that among the five types of cutting inserts evaluated (untextured, Titanium coated, dimple textured, parallel groove textured, and perpendicular grooved textured), the parallel groove textured insert performed the best exhibitingthe lowest surface roughness and tool wear.

IMPACT OF AGILE PRACTICES ON SOFTWARE DEVELOPMENT SUCCESS: A STRUCTURAL EQUATION MODELING APPROACH

2025-03-17T10:52:05+02:00

The purpose of this study was basically to examine how the Agile methodologies affect the success of the software development work by taking into consideration the mediating roles of team collaboration and communication efficiency. Structural Equation Modeling (SEM) in Smart PLS was applied to quantitatively analyze the data obtained from 300 software professionals in variety of industries using a quantitative research design. The study analyzed agile frameworks including Scrum, Kanban, and Extreme Programming (XP), and asses their effects on software quality and project success. It found that Agile methodologies not only made quintessential enhancements to the quality of the software and the success of the project but also improved efficiency of communication and promoted better collaboration between team members. Furthermore, I find both software development success and the effect of Agile on the Software Development Success to be mediated by the team collaboration and the team communication efficiency. It provides clues to the effective use of Agile implementation strategies by software firms as well as project management strategies.

Efficient ML Technique for Brain Tumor Segmentation, and Detection, based on MRI Scans Using Convolutional Neural Networks (CNNs)

2025-03-14T07:08:57+02:00

Experts need accurate segmentation and detection alongside the classification of Brain tumors from MRI images because this approach helps identify neurological problems early for timely treatment. Deep learning technology has made Convolutional Neural Networks (CNNs) effective in analyzing complex medical imaging challenges by developing automatic abilities to detect and categorize complex data features. This study used 1,251 Brain Tumor MRI images from BraTS2021 for model testing of CaPTk, 2DVNet, EnsembleNets, and ResNet50 towards brain tumor segmentation. The research utilized the DSC and HD metrics for its evaluation process. Importantly, EnsembleUNets achieved the minimum HD of 18 while reaching the maximum DSC of 0.92. The analysis of the radiomic feature confirmed that EnsembleUNets delivered the best CCC value at 0.75 together with the lowest RMSE at 0.52 and the highest TDI at 1.9 for tumor segmentation and classification in clinical practice. These findings show EnsembleUNets effectively perform brain tumor segmentation and classification and identification so healthcare professionals now have more effective guidance about implementing CNN-based clinical applications.

Keywords: Cyber-physical systems (CPSs). Security Protocols, Encryption, OpenVPN, IKEv2/IPsec, WireGuard, Quantum Computing

A Deep Learning-Based Enhanced Sentiment Classification and Consistency Analysis of Queries and Results in Search Using Oracle Hybrid Feature Extraction

2025-03-10T18:44:13+02:00

Multiple assessments of customer feedback play a vital role in the industry because they help enhance product quality while spotting major network issues and creating improved customer-facing services. A traditional sentiment analysis process relies on external machine learning frameworks that result in system integration issues and performance reduction because large data volumes transfer between different options such as API and FTP file sharing before applying data machine learning models that extract insights. This paper proposed a Machine learning-based Model based on (CNN, RNN, and DT Classifier) and focuses on extracting sentiment metadata that links to the user-selected topic or entity together with their search results. The real-time sentiment analysis system which operates in Oracle Autonomous Database uses OML4SQL and OML4PY components from Oracle Machine Learning to process Communication customer feedback obtained through the web-based Oracle APEX system. The predictive model developed by the research utilizes CNNs and RNN algorithms provided by Oracle to identify whether customer reviews are positive, negative, or neutral. After receiving training the model functions to classify fresh feedback immediately while bypassing dependencies on external AI platforms. The implementation occurs inside the Autonomous Oracle Database while bypassing API or FTP file-sharing methods. The analysis reveals OML4SQL and OML4PY succeed in customer sentiment analysis thus enabling Software organizations to acquire valuable business information for better service delivery and strategic choices. The findings from this research demonstrate that the Convolutional Neural Network (CNN) achieved the highest accuracy (92.5%), followed by RNN (90.2%), while DT (85.4%) performed relatively lower. The analysis of 15,500 customer reviews revealed that 48.1% were positive, 39.4% were negative, and 33.7% were neutral. Oracle machine learning tools (ML4SQL and OML4PY) provide real-time text analytics in Software databases which enables service-based decisions through automated sentiment analysis technology within customer support operations.

Keywords: Sentiment Analysis, Oracle Machine Learning for SQL (OML4SQL), Oracle Machine Learning for Python (OML4PY), Machine Learning.

Scrum-Driven Quality Improvement: Mitigating Software Failures Effectively

2025-03-07T11:29:25+02:00

The pressure on developmental teams increases due to the technical advancement of 21^st century, accomplishing the projects within specific time frame. Various Incremental development process initiated to achieve the growing customers’ demands. But it is a challenge for developers to opt the best methodology for quality enhancement of Software failure. During this process risk management, right product, time limitation and error rate should be kept in mind to avoid the risk of failure. In this study, one of the most famous agile methodologies, Scrum, was introduced to analyze the improvement of software failures. Through this analysis, we can suggest the implementation of Scrum for quality enhancement of software failures. For further study, researchers can work on Scrum flexibility and use of scrum model in all kind of software; also can work on scrum architecture to improve the performance and efficiency in software development management.

Enhancing Chronic Kidney Disease Prediction with a Hybrid Hard Voting Classification Approach

2025-03-14T01:40:57+02:00

Chronic kidney disease (CKD) affects over 10% of the global population, equating to more than 800 million people. It is prevalent among older adults, women, minorities, and individuals with diabetes and hypertension. CKD also presents a significant health burden in low- and middle-income countries. The growing prevalence of CKD and its adverse impacts emphasize the urgent need for enhanced prevention and treatment strategies. Machine learning (ML), a prominent application of artificial intelligence, has made significant strides in healthcare research. The aim of this study is to design an ensemble method for more accurate prediction of CKD in patients. The proposed ensemble hard voting classifier integrates four machine learning algorithms—Support Vector Machine, Logistic Regression, K-Nearest Neighbors Classifier, and Random Forest Classifier. The model is trained on a dataset sourced from Kaggle, which includes data from 400 patients with 25 features.

Keywords: Chronic kidney disease, machine learning, ensemble method, hard voting classifier.

Current State Analysis of Counterfeit Medicine Detection in Pharmaceutical Supply Chains in Pakistan

2025-03-11T10:27:40+02:00

Counterfeit medicines are a serious challenge in Pakistan's pharmaceutical supply chain, posing a serious threat to public health and economic stability, as well as the credibility of the health care system as a whole. This paper discusses the current status of counterfeit medicine detection mechanisms in Pakistan by using qualitative methods like an interview with industry leadership. The study deals with the examination of the present legacy systems; prevalence and effect of counterfeit medicines; and a gap in rules and technology that needs to be filled. Urgent need: modern technological integration, updated regulation frameworks, strategic solutions to overcome the growing danger of counterfeit drugs.

Exploring IoT Secuirty, Privacy and Data Protection

2025-03-09T18:14:44+02:00

Internet of things rapid growth is causing security and privacy concerns. Internet of medical things are devices designed for remote monitoring for the patients with chronic diseases. The security of the patient data at the collection, transmission and storage is critical for data integrity and privacy. The access to the patient data is limited only to users with high privileges i.e. for doctor and patient. On the other hand encrypted data is decrypted with shared keys by authorize user only. The technique is well suited for the access of data to only authentic users. The smart device connectivity with other devices is frequent that is sometimes raise questions related to security of these devices. The authentication of the devices is an important concern which is proposed this article. The authentication technique is robust as compared to other proposed technique due to having secure authentication at each step of devices to connect with main smart device. The Lightweight neural network method introduced for the network intrusion detection which is designed for IoT devices having low computing capability and low storage space. LNN model is excellent for classifying the smart devices traffic for normal and attack scenarios. In industrial IoTs trust management is the key factor to allow or deny the specific device to enter into the smart network. The trust between two nodes is calculated based on the device previous history with other devices. The approach is efficient for trust management based access to the devices. Fog assisted IoTs based on trust management and security component shows that their lightweight efficiency is higher as compared to other models for the security of resource constrained devices. For Industrial IoTs advanced encryption is best scenario where number of the devices frequently join and leave the network. The security measures at each level of the smart device is critical to security of data and privacy protection. The access based on human stored biodata or biometrics is secured enough for access to smart devices. The doubled encryption-decryption approach providing strong security measures for smart device generated data and users privacy. The Quick Response or QR code enabled technique providing critical information for device security enabled techniques increase buyer’s trust, which is important for deployment of security enabled smart devices around the world.

Keywords:&nbspInternet of things security, data protection, privacy, data encryption, network intrusion detection, fog-assisted IoTs, Trust management.

Exploring IoT Secuirty, Privacy and Data Protection

2025-03-08T16:48:38+02:00

Keywords:&nbspInternet of things security, data protection, privacy, data encryption, network intrusion detection, fog-assisted IoTs, Trust management.

A Comprehensive Analysis of Mechanical Strength and Durability in Cellular Light Concrete Blocks Modified with Multiple Additives

2025-03-13T12:26:30+02:00

The multipurpose material, cellular lightweight concrete (CLWC), comprises cement, fly ash, and a foaming agent. The popularity of cellular lightweight concrete can be attributed to its low weight, which lowers the structure's self-weight. With its cementitious qualities and mechanically entrained foam in the cement-based slurry, cellular lightweight concrete (CLWC) is a novel material that has significantly increased in popularity in the construction industry over the past ten years. CLC blocks have poor strength issues while being lightweight concrete with strong water absorption and thermal insulation qualities. CLC blocks are altered with additives like fly ash, silica fume, and marble dust to combat this issue. Due to its low strength, a study reveals that silica fume and marble dust have been utilized in several experiments to increase stability. The samples will undergo compressive strength, thermal conductivity, and water absorption tests. Regarding compressive strength, 3% silica fume and 7% marble dust work well. These findings are based on seven and 28-day curing times, while 14-day curing may be investigated in the future.

Keywords: Concrete blocks, Silica fume, foaming agent, Marble dust, Fly ash

ENHANCING THERMAL AND COMBUSTION PERFORMANCE OF GAS WATER TANK GEYSER THROUGH BAFFLE INTEGRATION

2025-03-17T11:07:28+02:00

Enhancing the thermal and combustion performance of gas water tank geysers is crucial for improving energy efficiency and minimizing fuel consumption. This study systematically investigates the impact of seven distinct baffle configurations baseline (no baffle), strip, cylindrical, conical, finned conical, frustum, and bladed frustum on the thermal and combustion performance of a gas water tank geyser. Experimental findings demonstrate that structured baffle designs significantly enhance heat transfer, optimize fuel-air mixing, and reduce fuel consumption. Among the tested configurations, the bladed frustum baffle exhibited the highest thermal efficiency (69.03%), surpassing the conical baffle (61.24%) by 7.79% while achieving an 8.4% reduction in gas consumption. The frustum baffle demonstrated a 5.83% increase in thermal efficiency (67.07%) and a 6.49% reduction in gas consumption compared to the conical baffle, underscoring its superior heat retention and transfer capabilities. In terms of combustion performance, the frustum baffle achieved a 3.99% higher combustion efficiency (69%) than the conical baffle (65.01%), ensuring more complete fuel utilization. However, the bladed frustum baffle showed a 0.61% decline in combustion efficiency (64.4%), attributed to increased stack temperature and altered combustion dynamics. Overall, the frustum baffle emerges as the optimal configuration, offering the best balance between improved thermal efficiency (67.07%), stable combustion (69%), and reduced fuel consumption. These findings provide a robust framework for enhancing the energy efficiency of gas-fired heating systems in residential applications.

IN-SITU ELEMENTAL DETECTION IN TOBACCO AND ASH BY LASER INDUCED BREAKDOWN SPECTROSCOPY: IMPLICATIONS FOR HUMAN HEALTH AND ENVIRONMENTAL SUSTAINABILITY

2025-03-18T10:38:54+02:00

Laser induced breakdown spectroscopy (LIBS) is a versatile technique that is used to determine elemental composition in different samples. It is simple and fast multi-elemental analysis technique which provides potential tool in situ chemical analysis with high resolution, better limit of detection (LOD) and is cost efficient. Smoking tobacco in cigarettes amplifies the risk of growing certain diseases such as cancer, heart disease, stroke, lung diseases, diabetes, and chronic obstructive pulmonary disease (COPD), which includes emphysema and chronic bronchitis. About 80% of lung cancer as well as about 80% of all lung cancer deaths are due to smoking. Smoking tobacco is a considerable source of consuming several harmful and toxic elements that are unhealthy for human body. The presence of some toxic elements causes serious health concern to active as well as to passive smokers. In the present study, we employ LIBS technique to identify and detect toxic and harmful elements in tobacco and ash from four popular Pakistani cigarette brands, ensuring human health protection, agricultural safety, and environmental sustainability. The Q-switched ND: YAG (neodymium-doped yttrium aluminum garnet) laser (λ=1064 nm) with laser energy 90 mJ and pulse duration 10 ns were used to ablate the samples. From the recorded optical emission spectra of these samples several elements were detected (Fe, Ca, Mn, Sc, Ti, Cr, Sc, Sr and Ni) among which Chromium Cr, Nickel Ni and Strontium Sr are highly toxic elements. Furthermore, the presence of toxic and heavy metals in ash could be significant contributor to metal load in soil as well as for human body. We calculated the electron temperature (Te) by the Boltzmann Plot Method from the spectroscopic analysis of the transition lines of Fe-I (iron). Stark Broadening Method was used to determine electron number density (Ne) from the transition line of Fe-I at 422.413 nm and 649.25 nm for tobacco and ash sample respectively, under the assumption of local thermodynamic equilibrium. The present results demonstrate that LIBS is a powerful diagnostic tool for effectively tracing harmful and toxic elements in various solid samples, facilitating the development of protective environmental measures.

OVERCOMING ENERGY EFFICIENCY CHALLENGES IN C-RAN: A REVIEW OF RECENT ADVANCES AND SOLUTIONS

2025-03-19T08:09:40+02:00

Cloud Radio Access Networks (C-RAN) is one of the novel approaches to cell network architecture by centralizing baseband processing devices (BBUs) into cloud statistics facilities and dispensing far off radio heads (RRHs) at mobile web sites. While C-RAN supports aspects of higher scalability, flexibility and network management, it also creates significant large power-intake demanding scenarios due to its operation with multiple RRHs and the power-in-depth BBU pools. This paper examines the major challenges in realizing power efficiency in C-RAN with dynamic site traffic loads, idle power consumption as well as cooling requirements in central data centers. For example, we consider several alternative designs such as dynamic BBU pooling, sleep mode techniques for RRHs, AI-based visitors prediction, and integration of renewable sources of energy. Hence, based on such results, it can be inferred that strength consumption decreases significantly without degrading the community performance with such technologies, especially by implementing them at various layers of the network. This work concludes with some future research guideline on how to improve the energy efficiency of C-RAN in the scenario of 5G and previous networks.

INVESTIGATING OF BLACK HOLE ATTACK IN WIRELESS SENSOR NETWORK

2025-03-20T09:03:15+02:00

As we enter the era of sensors different types of sensors are deployed everywhere for making different types of Networks such as MANET, WBAN, VANET, etc. All of these are the categories of wireless sensor networks and in all categories, the sensor is deployed to collect data and transmit it to the sensor head node or base stations during the data transmission a variety of security threats are available some of them are active and some of them are passive all active. In this paper we discussed various security attacks that directly affect the performance of a node or network (performance-oriented attack), creating the violence in layers (layer-oriented attack) and achieving some goals (goals-oriented attack) in detail also discussed the countermeasures for all types of attacks. This paper is specially designed for the investigation of black hole attacks in the WSNs network we also cover the detection of blackhole attacks and their prevention technique.

ACCESS CONTROL MODEL FOR DATA STORED ON CLOUD COMPUTING

2025-03-20T09:18:57+02:00

Cloud computing can be referred as technology based on internet, having shared, adaptable foundation that can be utilized as administration by clients. Compositely cloud computing is software and hardware being conveyed by internet as a service. It is an extraordinary advancement innovation getting well known because of minimal effort, adaptability and versatility according to client’s necessity. The objectives of this research were to concentrate on client’s data security in cloud computing i.e. data storage security issues and how to limit unapproved access to information by proposing access control model. Accessible arrangements were introduced by the researcher and then an access control model was recommended. Many access model were accessible yet those model do not satisfy the security necessities as per service providers and cloud is always under assaults of hackers and data integrity, accessibility and protection were traded off. This research presented a model keeping in view the requirements of service providers that upgraded the information security in terms of integrity, accessibility and protection. The proposed model helped the hesitant clients to effectively choose to move on cloud while understanding the dangers related with cloud computing.

AI-POWERED ANOMALY DETECTION IN SOFTWARE LOGS: A MACHINE LEARNING APPROACH FOR PROACTIVE FAULT DIAGNOSIS AND SELF-HEALING SYSTEMS

2025-03-20T09:27:47+02:00

Due to the complexity of modern software systems the amount of logs generated to assist with monitoring and fault diagnosing has become way too large for manual processing. This paper aims at developing the architecture for identifying anomalous patterns in the software log files through the application of advanced machine learning and deep learning algorithms towards fault diagnosis for self- healing systems. Traditional rule based approaches cannot fit the modern complex scenarios as well as the large amounts of data that are produced in the form of logs. Machine learning approaches, including deep learning structures like LSTMs and Transforms, are more effective at detecting anomalies due to their ability to capture contextual dependencies inherent in log sequences. It also offers resolutions to the problem of a scarcity of labeled data in the utilization of self-supervised learning approaches, along with contrastive learning. Additionally, self-C damaged control mechanisms based on reinforcement learning as well as a rule and based automation recurrently correct faults decreasing the non-availability of the system. Several models are assessed on log datasets with different evaluation metrics such as precision, recall, F1-score, and AUC-ROC. The results when testing suggest that Transformer-based models yield the best performance as compared to other conventional machine learning methods while at the same time requiring more computational resources. Self-healing systems cut down on downtime by as much as 68.2 percent; such characteristics make AI promising for strengthening system performance. That is why some issues, like model interpretability, high computation costs, and real-time processing, are still present. Mitigating these challenges by employing lightweight deep learning models, explainable AI methods, and the ability to deploy these algorithms at scale will be instrumental in advancing the use of AI-based anomaly detection and self-healing systems in safety- critical software applications. This work presents a state-of-the-art review of AI- based log anomaly detection methods and discusses potential research directions for improving scalability, interpretability, and practicality in real-world applications.

STEGANOGRAPHIC SECRET COMMUNICATION USING RGB PIXEL ENCODING AND CRYPTOGRAPHIC SECURITY

2025-03-21T09:35:49+02:00

This research presents an advanced and highly practical strategy for secure communication through steganography, leveraging RGB pixel encoding and cryptographic security. Our novel methodology segments message data into 3-bit, 3- bit, and 2-bit components, which are seamlessly integrated with the corresponding RGB pixel values of a digital image. This structured embedding process generates a logarithmic (LOG) table, which is subsequently encrypted using the Rijndael Managed algorithm, enhancing both confidentiality and resistance to cryptographic attacks. Comparative analysis demonstrates that our approach maintains image quality with less than a 1% variation in pixel intensity, significantly outperforming traditional steganographic techniques such as the Least Significant Bit (LSB) method. The proposed method establishes a robust and imperceptible communication channel, offering superior security, undetectability, and resilience against steganalysis.

READABLE TEXT RETRIEVAL FROM NOISE-INFLUENCED DOCUMENTS USING IMAGE RESTORATION METHODS

2025-03-21T09:43:50+02:00

Documents scanning has become a necessary phase in official record keeping of everyday business environment. Typically, scanned document images in digitized format suffer from various types of noise which create serious problems at document reading time. This noise may be due to several reasons low quality paper, paper aging, scanner assembly and tonner, unskilled machine operator, or due to some copying machine artifacts. The removal or elimination of noise in scanned documents is still a big challenge for researchers in the digital era. Already performed work on digitized handwritten, and machine-printed degraded historical documents, but we have experimented with different datasets such as the Media Team Document Database manually scanned noisy documents, and decided to use, a collection of scanned noise-affected documents, which are available on the websites. We have transformed the noise-influenced image document into a binarized document. After this, we applied noise reduction techniques for textual data enhancement so that the text would be in readable and noise-free form. An Adaptive Gaussian Mixture Model based on Expectation Maximization (EM) has been used to restore the image pixels, with the values expected to be the original ones. The enhanced text in its visual aspect and improved quantitatively measured parameters show the restored documents. We have calculated Signal to Noise Ratio (SNR), Mean Square Error / Mean Square Difference (MSE/MSD), Peak Signal Noise Ratio (PSNR), Contrast, and Energy for quantitative parameters to evaluate the performance measures of the proposed method of document restoration comparative to the state-of-the-art methods. Our research is quantitative, as we have performed experiments on digital sensor data and the evaluation of the results based on computational techniques. Our results are successful, support the proposed methodology, and perform well in comparison to the state-of-the-art methods. Overall, the proposed methodology is easy to understand and simple to implement.

USING GENERATIVE AI FOR SIMULATING CYBER SECURITY ATTACKS AND DEFENSE MECHANISMS: A NEW APPROACH TO AI-DRIVEN CYBER THREAT MODELING

2025-03-22T10:10:41+02:00

Awareness of cyber risk is a crucial factor in today’s world, where new forms of threats are potentially dangerous to companies’ IT frameworks. The reaction-based detection, particularly the signature based detection and the rule based systems have their limitations in combating new and advanced attacks because they are deficient in the proactive approach. With threats emerging in the cyber realm in a dynamic way, it is imperative that organizations seek out active and conscious defenses. This paper aims at bringing forward the possibility of using Generative Artificial Intelligence (AI) in the process of modeling cyberattacks and defense strategies: a new approach to Defending AI Cyber Threats. Generative AI models such as GAN and VAE are used to generate realistic attack scenarios that are useful in assessing and improving security systems. These allow the emulation of new scenarios that have not yet been experienced in cyberspace that, in turn, can be used to assess defence mechanisms. This is why this approach is popular in cybersecurity research since it can represent such threats as APTs and other types of threats that may appear over time while advancing through the specific attack life cycle. Through leveraging generative AI for threat modeling, organizations start shifting from a reactive approach to threat modeling, with security features matching up with the advancing complexity of threats.

COMPREHENSIVE ANALYSIS OF FRAUD DETECTION PREVENTION SYSTEMS FOR ACCURACY AND EFFICACY

2025-03-25T08:41:58+02:00

Financial fraud, waste, and abuse cost global economies an estimated $5.4 trillion annually, with digital payment platforms experiencing unprecedented vulnerability. This study presents a systematic evaluation of contemporary fraud detection and prevention systems across major financial institutions, analyzing their accuracy, efficacy, and scalability in high-volume transaction environments. The mixed-methods approach combined quantitative performance metrics from financial institutions with qualitative assessments from cybersecurity specialists to evaluate detection algorithms across four dimensions: detection accuracy (false positive/negative rates), computational efficiency, adaptability to emerging threats, and implementation feasibility. Results demonstrate that hybrid approaches combining supervised machine learning with unsupervised anomaly detection achieved superior performance (92.7% detection accuracy) compared to traditional rule-based systems (78.3%). Notably, models integrating graph-based network analysis with deep learning techniques showed particular promise in identifying sophisticated organized fraud schemes, reducing false positives by 34% while increasing true positive rates by 27% compared to standalone approaches. The rise of cloud computing and mobile transactions has fundamentally altered the fraud landscape, requiring detection systems that can process and analyze real- time streaming data at unprecedented scale. The comprehensive classification framework categorizes existing detection systems based on algorithmic approach, fraud typology, and quantitative performance metrics across diverse financial contexts. The study identify critical challenges in current implementations, including the increasing sophistication of adversarial attacks, computational constraints in real-time environments, and the dynamic nature of fraudulent behaviors. Based on our findings, we propose a next-generation architectural framework for financial fraud detection that emphasizes real-time adaptability, explainable AI components, and cross-institutional collaboration, potentially reducing overall fraud losses by an estimated 41% when implemented at scale.

OPTIMIZING PREFABRICATED BUILDING INDUSTRY INTEGRATION: A STUDY OF CONSTRUCTION THROUGH INDUSTRIAL SUPPLY CHAIN DYNAMICS AND UPSTREAM/DOWNSTREAM ENTERPRISE ACCESS STANDARDS

2025-03-25T08:54:50+02:00

This study examines the supply chain integration in the prefabricated building industry, focusing on the relationship between upstream and downstream enterprises. The research identifies critical risk factors across the project lifecycle using an Importance-Performance Analysis. Key findings reveal that major risks like high costs and policy changes, are more prevalent in the early project stages, while risk management is weakest in later stages, particularly in manufacturing and construction. The study highlights challenges in standardization, quality control, and technology adoption, emphasizing the importance of government policies and industry standards in shaping supply chain integration. Strategies for enhancing integration are proposed, including the development of industry-wide standards, investment in skills training, and adopting a lifecycle approach to project management.

ADVANCES OF MACHINE LEARNING: A SURVEY OF METHODS, BENCHMARKS, MODELS AND DATASETS IN INDUSTRY APPLICATIONS

2025-03-26T10:01:24+02:00

This study explores the fundamentals of Machine Learning (ML), a sub-discipline of Artificial Intelligence (AI) that enables systems to learn and make decisions based on new data without explicit programming. It provides an introduction to the various types and approaches to ML, highlighting the models' ability to improve over time as they process more data. Additionally, the paper traces the historical evolution of ML, from statistical learning theory to neural networks, and examines its relevance in modern society, driven by the availability of big data and advances in computational power. Furthermore, the study investigates the application of ML across major industries such as healthcare, finance, and retail, demonstrating its potential to solve complex problems, enhance decision-making processes, and transform industries.

CLOUD COMPUTING ADOPTION IN SMES: AN EMPIRICAL STUDY USING PLS-SEM"

2025-03-26T10:06:44+02:00

Cloud computing is game changing technology for these small and medium-sized enterprises (SMEs) as they can now avail cheap and scalable IT solutions. This study examines the relationship from perceived usefulness, security concern and cost efficiency to SMEs cloud adoption, and used organization readiness as the mediator in the process. Using a quantitative explanatory research design, data were collected through a 7-point Likert-type scale from a sample of 209information technology (IT) managers, system administrators, and decision-makers in SMEs of Karachi. A Structural Equation Modeling (SEM) was applied on the data using Smart PLS, which explored both direct and indirect relationships. It found that both perceived usefulness and cost efficiency had a significant impact on cloud adoption, while security concerns had a significant deterrent effect. Moreover, the finding demonstrated that organizational readiness has a significant mediating role about the relationships between the independent variables and cloud access. And the findings of the study can support the decision-makers in the SME sector, as well as policymakers with investment towards cloud technology sentiment, especially IT infrastructure improvements and more managerial involvement in cloud adoption.

THE DESIGN OF HIGH PERFORMANCE PHOTO-TRANSISTOR BASED ON SEMICONDUCTOR NANO-MATERIALS

2025-03-27T11:13:50+02:00

Background: Phototransistors are key components in modern optoelectronics due to their high sensitivity and wide spectral response. However, enhancing performance metrics such as responsivity, detectivity and response time remains a significant challenge. Semiconductor nanomaterials offer promising solutions owing to their tunable band gaps, high carrier mobility and strong light-matter interaction. Objective: This study aimed to design and fabricate high- performance phototransistors using GaN-ZnO solid solution nanowires and WS₂– InGaZnO heterojunctions and to evaluate the impact of band gap engineering, synthesis parameters, plasmonic enhancement and heterostructure formation on device performance. Methods: GaN-ZnO nanowires with varying ZnO concentrations were synthesized and deposited onto Si/SiO₂ substrates to form the phototransistor channel. Device performance was evaluated under different synthesis temperatures and durations. Ag nanoparticles were introduced for plasmonic enhancement. Separately, WS₂–InGaZnO heterojunctions were fabricated via CVD and sputtering techniques. Phototransistor architecture employed bottom-gate configuration with Ti/Au source-drain electrodes. Key parameters such as photocurrent, responsivity, detectivity, response time and operational stability were analyzed. Results: Band gap tuning from 3.4 eV (GaN) to 2.6 eV (Zn-rich nanowires) enhanced visible light absorption. Optimal synthesis at 850 °C yielded highest responsivity (95.3 A/W) and detectivity (2.1 × 10¹¹ Jones). Ag nanoparticle decoration further improved responsivity to 131.7 A/W and reduced response time to 5.9 ms. The WS₂–InGaZnO heterojunction device exhibited superior performance with responsivity of 122.5 A/W, detectivity of 3.8 × 10¹¹ Jones, and excellent stability (91.7% retention over 50 cycles). Conclusion: The integration of band gap-engineered nanowires, plasmonic enhancements and heterojunction structures significantly advances phototransistor performance. These findings provide a strong foundation for developing next-generation, broadband and high-sensitivity photodetectors for practical optoelectronic applications.

LEVERAGING AI TO IDENTIFY ANOMALIES IN ELECTRICAL SYSTEMS AND COMMUNICATION NETWORKS, SAFEGUARDING CRITICAL INFRASTRUCTURE AGAINST CYBER-ATTACKS

2025-03-27T11:41:43+02:00

Every year the dependency on essential infrastructure systems keeps growing thus making them more vulnerable to complex cyber-attacks. The contemporary and multifaceted terrorist threats can rarely be handled effectively through conventional cybersecurity protocols. The paper evaluates how Artificial Intelligence (AI) coupled with Machine Learning (ML) performs forecasting and mitigation of security threats which affect critical infrastructure systems. Machine Learning together with Artificial Intelligence methods perform exceptionally well in real-time extensive data processing functions while discovering patterns that indicate cyber-attack occurrences. The research investigates sophisticated AI and ML algorithms including supervised learning and unsupervised learning as well as reinforcement learning because they demonstrate their ability to detect intrusions and assess vulnerabilities and predict threats. Primary analysis evaluates essential hurdles that involve AI model adversarial attacks as well as privacy concerns and data requirements for high-quality training samples. This research evaluates XAI as a key factor for enhancing transparency and reliability in security systems built using ML so they can be properly deployed in critical infrastructure environments. The paper evaluates this integration process between threat intelligence systems based on AI with present-day cyber frameworks while focusing on both real-time threat mitigation and flexible responses toward shifting attack routes. A.I.-driven problem recognition systems within Industrial Control Systems (ICS) demonstrate their ability to shorten operational stoppages while reducing financial damage and preventing infrastructure breakdowns according to a studied implementation. The outcomes demonstrate AI together with ML can boost critical infrastructure's cyber-attack resilience therefore establishing proactive cybersecurity practices instead of reactive ones. The study finishes by outlining projects for future research to create robust AI algorithms that defend against attacks along with standardized data accumulations for crucial infrastructure defense and strengthened cooperation between experts from different fields involving AI and cybersecurity specialists and managers of critical infrastructure systems. Vital infrastructure cybersecurity can achieve significant enhancement through the use of AI and ML technologies which will maintain key services dependable and safe and operational.

NEXT-GENERATION BIODEGRADABLE POLYMERS: INNOVATIONS FOR A GREENER PLANET

2025-03-28T08:57:02+02:00

The global demand for sustainable materials has driven significant advancements in next-generation biodegradable polymers, offering promising solutions for reducing environmental pollution. This study explores recent innovations in biodegradable polymer development, emphasizing eco-friendly synthesis, enhanced performance, and diverse applications. Modern approaches focus on utilizing renewable resources such as plant-derived polysaccharides, proteins, and microbial biopolymers to create materials with tailored mechanical properties and controlled degradation rates. The integration of nanotechnology and bio-based additives has further enhanced polymer stability, strength, and biodegradability, making them suitable for packaging, agriculture, medicine, and textiles. Additionally, advancements in polymer processing techniques, including 3D printing and green chemistry methods, have facilitated scalable and energy-efficient production. These next-generation polymers not only minimize reliance on fossil fuels but also reduce plastic waste accumulation through accelerated degradation in natural environments. Challenges such as cost, production scalability, and end-of-life management remain, but ongoing research focuses on optimizing performance while ensuring environmental compatibility. The paper highlights emerging trends, lifecycle assessments, and future perspectives for integrating these materials into circular economies. Ultimately, next-generation biodegradable polymers represent a transformative shift towards a more sustainable and resilient planet, addressing the urgent need for eco-conscious material innovations.

AN EFFICIENT METHOD TO LOCATE LICENSE PLATES UNDER DIVERSE ENVIRONMENTS

2025-03-28T10:06:46+02:00

Advances in artificial intelligence (AI) have made car and road object identification more efficient. Researchers face a difficult problem in detecting license plates (LPD), which necessitates the use of a dependable and precise automatic detection method. Recent approaches to deep learning have showed promise, although they are frequently limited to specific locales or private datasets. This study focuses on accurately finding license plate locations using machine vision and deep learning approaches, especially under adverse weather situations such as blurriness, nighttime, and high brightness. A modified RCNN model is developed for LPD, which uses the China City Parking Dataset (CCPD) to forecast nine different stages. The suggested method includes data preprocessing, weight generation, and model training. The model is taught using self-generated weights rather than pre-learned weights, and it detects number plates with 98% accuracy under a variety of weather situations. After 50 epochs of training, the detection module predicts suitable bounding boxes. The Region Convolutional Neural Network (RCNN) is highly effective in detecting number plates under diverse environments.

Federated Learning for Distributed Anomaly Detection in Network Traffic Using GRU-Based Models

2025-03-31T11:55:09+03:00

In this work, we present a novel machine learning method for anomaly detection in network traffic based on GRU based federated learning. Our decentralized method is supported by extensive experimental results and comparisons with existing techniques, and successfully addresses scenarios where centralized servers are not feasible due to privacy concerns or other constraints, and successfully detects anomalies in the distributed environments..

A NEW REAL-WORLD HAZY IMAGE DATASET FOR IMAGE ENHANCEMENT AND RECOGNITION

2025-04-02T09:58:06+03:00

Many computer vision tasks are benefited significantly from deep learning. However, research in the field of single image dehazing is still needed. The unclear image boundaries and dense haze degrade the visible quality of haze-free images. For measuring the efficiency of techniques, many deep learning-based approaches are evaluated on the real-world hazy image datasets for more credibility. Moreover, this study discusses on creation of a dataset that includes real-world hazy images. The dataset is valuable for computer vision and image processing research. It offers training resource for deep neural networks and machine learning models to handle more hazy circumstances.