Iterated Local Search (ILS) is a well-established metaheuristic that has been widely applied to combinatorial optimization problems owing to its balance between solution diversification and intensification. This work emphasizes ILS for the scope of the Asymmetrical Traveling Salesman Problem (ATSP) and the Asymmetrical Generalized Traveling Salesman Problem (AGTSP) respectively, which accounts for the directionally dependent costs of routes and cluster of nodes recognized in the routing difficulties of real-world applications. The experiment also assesses whether ILS can explore difficult search regions while also conserving solution quality and avoiding rapid convergence. This study examines the feasibility of ILS when using TSPLIB benchmark instances subject to directed route costs, time windows and load capacity constraints, while including a realistic routing network, including depots, customers and stops, alongside an examination of the systematic conversion of symmetric TSP instances into asymmetric representations to accurately capture directionally dependent travel costs. The results demonstrate that ILS can produce high quality solutions to both the ATSP and AGTSP under conditions of increasingly complicated routing. Future research will be directed towards improving ILS with a hybrid metaheuristic framework and subject to large scale logistics datasets to improve viability and scale capability.
This paper presents a small, scalable, and affordable solar-powered inverter system, designed to provide a continuous power supply in off-grid settings and during grid outages worldwide. The proposed system utilizes lithium-ion batteries charged by solar energy to power a high-efficient and low power dissipation inverter constructed using durable power transistors, a step-up transformer, and a custom-designed filter circuit. The inverter generates a near-sinusoidal AC output capable of supporting a wide range of loads, thereby reducing dependency on fuel-generated grid electricity, particularly in underserved regions. The system emphasizes cost-free energy storage and reducing energy losses within components. It maintains a stable near-sinusoidal voltage under practical load conditions. The modular design integrates energy conversion, storage, and AC power delivery into a unified system. With a Total Harmonic Distortion (THD) of only 2.17%, the inverter delivers high-quality voltage output with a near-sinusoidal waveshape at lower complexity compared to conventional systems. The system is suitable for mobile applications, emergency supply of power during outages, disaster relief, and rural electrification. The proposed solution demonstrates the accessibility and effectiveness of solar technology in practical, real-world applications.
In this work, various stationary objects were detected within a room measuring 4.88 × 3.91 m². The data were captured by a 24 GHz commercial radar module, OPS243-C. The experiments were performed in two scenarios: over each wall or at different angles of the radar positions, and, secondly, by mounting this radar module on a rotating motor to scan each object over a 360° coverage. In addition, to suppress background noise and filter out strong signals in the heatmap results, a noise threshold and a Gaussian smoothing filter were applied. The predicted distances of objects were then obtained by applying suitable curve-fitting models to radar data captured in each scenario. The corresponding images captured in each scenario were used to validate the 2D and 3D heatmaps for object distances after curve fitting. Thus, for each wall scan, a linear curve-fitting model was found to be sufficient, whereas for entire-room scanning, a 6th-order polynomial curve-fitting model was found to provide the best predicted distances by the module. The methodology discussed in this work is useful for analyzing field-grown crops and continuously monitoring their health.
The increasing complexity of electronic components and their stringent performance requirements necessitate precise and reliable testing methodologies. This study explores the development and implementation of a pin electronic system with level specification using Automated Test Equipment (ATE) testers. The purpose of ATE implementation is to drastically reduce human error, increase test throughput, and provide precise, real time data analysis. By utilizing advanced ATE platforms, the study evaluates the accuracy, repeatability, and scalability of the testing process. The proposed approach for the creation of the DTI board is to study the behavior of components that contribute to the levels of VIH, VIHH, and VIL values, thus, influencing the measurement of the PE electronic pins. The results demonstrate significant improvements in measurement precision and test cycle efficiency, offering a robust solution for quality assurance in semiconductor manufacturing. These findings provide valuable insights into optimizing test protocols for high-performance electronic components and pave the way for future advancements in ATE-driven testing methodologies.
In the modern digital environment, where vast amounts of multimedia data are continuously exchanged across global networks, ensuring secure and covert communication has become a critical challenge. Steganography plays a crucial role in this context by allowing the imperceptible embedding of secret information into digital media, including images, audio, and video signals. However, most state-of-the-art steganographic methods rely on computationally intensive mathematical transforms, including the Discrete Cosine Transform (DCT), Singular Value Decomposition, and various wavelet-based or deep-learning-based approaches. While these techniques achieve high robustness and embedding capacity, their high computational complexity limits practical implementation on resource-constrained platforms such as mobile devices, embedded systems, and Internet-of-Things environments. A promising direction for overcoming these limitations is the development of code-controlled steganographic methods, which operate directly in the spatial domain. Their efficiency is based on establishing deterministic relationships between DCT coefficients and those of the Walsh-Hadamard Transform (WHT), enabling precise code-based control of the embedding process. To date, such research has been limited to classical Sylvester-type Hadamard matrices. However, multiple non-equivalent classes of Hadamard matrices exist, and exploring their properties may unlock new opportunities for efficient and secure information embedding. This paper presents a comprehensive analysis of the correspondence between DCT coefficients and WHT coefficients derived from five non-equivalent classes of Hadamard matrices of order 16. Large-scale experimental research was performed using 530 images from the NRCS public database. Each image was processed in 16×16 blocks, with controlled perturbations applied to individual DCT coefficients. The resulting changes in WHT coefficients were analyzed to determine the mappings between the transform domains. The results reveal that while the classical Sylvester matrix preserves a one-to-one correspondence with DCT, non-classical Hadamard matrix constructions exhibit overlapping mappings, particularly in the high-frequency range. This overlap, although reducing spectral resolution, introduces beneficial redundancy for steganographic applications, enhancing robustness and resistance to steganalysis. The findings provide a foundation for developing high-performance, spatial-domain steganographic schemes that exploit the structural diversity of non-classical Hadamard matrices for improved efficiency and security.
LoRa, or Long Range, is an increasingly popular wireless communication technology for Internet of Things (IoT) applications, offering long-distance connectivity with low power consumption. Radio channel propagation plays a critical role in determining the performance of a LoRa system. This paper provides a comprehensive review of radio channel propagation in the context of LoRa communications. It introduces various factors influencing LoRa signal propagation and examines recent research efforts aimed at understanding the propagation characteristics of LoRa radio channels and addressing the associated challenges. In conclusion, selecting an appropriate propagation model for LoRa, the propagation model requires consideration of environmental variations ranging from urban to rural settings. The Free Space Path Loss model is suitable for open areas and short distances, while the Log-Distance Path Loss provides flexibility across various conditions. The Okumura-Hata model is well-suited for urban and suburban environments, the ITU-R P.1546 model offers broad coverage across a wide range of environmental types, and the SUI model presents a comprehensive solution to a wide range of conditions. The appropriate model selection depends on the deployment of location, frequency of operation, and specific environmental conditions in which LoRa is implemented.
Pneumonia is one of the leading causes of illness and death globally. If not treated promptly, it can be fatal. Early detection of pneumonia significantly reduces mortality rates and improves the chances of recovery. Among the key diagnostic tools for pneumonia is the chest X-ray, which is widely used due to its affordability. However, diagnosing pneumonia based on chest X-ray images can be challenging, as the visual symptoms may resemble those of other respiratory conditions. These diagnostic challenges are often subjective and dependent on the practitioner’s experience. To address this, computer-aided diagnostic (CAD) technologies can assist healthcare professionals in improving diagnostic accuracy. This research proposes a machine learning-based method to classify pneumonia using chest X-ray images. Specifically, it presents a framework that employs Random Forest (RF), K-Nearest Neighbor (KNN), and Support Vector Machine (SVM) models for automated pneumonia detection. This study involves developing and evaluating these models on chest X-ray images resized to 224 × 224 pixels. To address class imbalance, the Synthetic Minority Over-sampling Technique (SMOTE) was applied. The results demonstrate high classification accuracy: SVM achieved 97%, KNN 98% and RF 94%. These outcomes outperform some previously reviewed models and show potential for accelerating early diagnosis and treatment of pneumonia disease.
Wireless Sensor Networks (WSNs) are widely applied in various fields, such as healthcare, security, agriculture, and education, where they simplify complex tasks and enable advanced monitoring. However, effective deployment remains a critical challenge in WSNs, affecting coverage, connectivity, energy efficiency, and network lifespan. Despite the importance of deployment optimization, no comprehensive Systematic Literature Review (SLR) has been conducted on metaheuristic approaches to address this challenge. This paper systematically reviews the demographic distribution, state-of-the-art metaheuristic solutions, and experimental evaluation methods related to the WSN deployment problem. A total of 112 studies were analyzed to address key research questions. Our findings provide a structured overview of current methodologies and identify significant research gaps. Key challenges observed include limited availability of open datasets, over-reliance on simulation validation, and the lack of comprehensive performance metrics such as latency and delay. These insights suggest future research directions for further optimizing WSN deployment.
Machine learning has increasingly been applied to behavioral analytics, yet its potential in lifestyle classification remains underexplored. This study utilizes a Random Forest classifier to predict lifestyle categories based on behavioral patterns from the Half a Million Lifestyle Dataset. A key challenge in lifestyle classification is balancing accuracy and generalization, which was addressed through parameter optimization to mitigate overfitting. To assess real-world applicability, 93 students provided behavioral inputs, which were processed through a Python-based program. The model successfully classified participants into Fitness Enthusiast (41), Health-Conscious (50), Eco-Friendly (1), and Social Media Influencer (1) categories, achieving an accuracy of 75.07%. These results confirm that machine learning can effectively predict lifestyle behaviors, with implications for personalized health interventions and behavioral analytics. This study underscores the significance of parameter tuning and feature selection, offering a scalable and data-driven approach to behavioral classification and wellness management.
The rapid development of Internet of Things (IoT) revolutionized the method to secure and monitor our homes. As smart home technology becomes increasingly accessible, the integration of robust, scalable, and cost-effective security systems is no longer a luxury but a necessity. This study proposed the creation and development of a home security system integrated with Home Assistant and Tailscale VPN using the ESP32-CAM WROVER microcontroller. By leveraging the IoT ecosystem, the system features real-time door detection, intrusion alarms, and remote monitoring capabilities. A reed switch sensor detects door or window activity, while the ESP32-CAM WROVER captures and transmits images of potential intrusions to a Linux-based Hass.io server. Notifications and images are sent to the user’s smartphone for real-time updates, ensuring timely responses to security threats. The system is secured via a Tailscale VPN, allowing authorized users seamless access through MagicDNS. The research highlights the cost-effectiveness and efficiency of utilizing Linux-based virtual machines for smart home applications, as well as the performance advantages of the ESP32-CAM WROVER microcontroller. System testing performed by a single user with several activation attempts shows a success rate of ~93.33% for the alarm output, with a ~6.67% error rate attributed to rapid state changes or hardware noise. In contrast, the camera snapshot component demonstrated 100% reliability, with no errors in capturing or transmitting images. Comprehensive tests confirm the system's ability to enhance home security with its scalable and flexible architecture, making it a practical solution for modern smart homes.
Bias in AI-driven diagnostic models has raised serious concerns regarding fairness in healthcare delivery, particularly for chronic diseases like diabetes. This study investigates algorithmic bias in diabetes prediction models and evaluates the effectiveness of three fairness-aware approaches: Fairness-Aware Interpretable Modelling (FAIM), Fairness-Aware Machine Learning (FAML), and Fairness-Aware Oversampling (FAWOS). The same dataset and experimental setup were used to ensure a fair comparison across models. FAIM employs interpretable decision trees to enhance transparency but lacks explicit fairness mechanisms. FAML incorporates adversarial fairness constraints, achieving perfect fairness metrics while maintaining acceptable accuracy. FAWOS addresses class imbalance using SMOTE, improving overall classification accuracy without enforcing fairness. Results show that while each method has strengths, none independently achieves an optimal balance of accuracy, fairness, and interpretability. Therefore, this paper proposes a hybrid approach that integrates multiple bias mitigation strategies to support fairer and more reliable AI applications in clinical settings. This study contributes a structured comparative evaluation framework and offers actionable insights for the development of ethical AI models in healthcare diagnostics.
Conventional human-operated surveillance is prone to errors caused by distractions, fatigue, or biases. The proposed project seeks to improve the effectiveness and precision of surveillance by utilizing an image detection algorithm and the Internet of Things (IoT). This project overcomes the limitations of human detection by deploying a drone equipped with a camera that is able to identify people and suspicious individuals. The system is designed to identify individuals within the drone's visual range, evaluate the accuracy of person detection, and provide immediate monitoring via IoT connectivity. Person detection is performed using the You Only Look Once (YOLO) algorithm, and the project's scope is limited by the camera's quality, field of view, and resolution, which restrict its ability to recognize individuals within a certain range. Based on the results from a dataset of 3,500 images, the person detection algorithm achieves a mean average precision (mAP) value of 0.835, a confidence ratio of 0.79, and high accuracy in the confusion matrix. Performance is further improved by integrating OpenVINO for computers running on Intel CPUs. The IoT-based monitoring system is implemented using a Streamlit web-based application and Telegram application. The results show a 2-second delay in streaming from the Raspberry Pi to the host computer, with an average speed of 100 ms per frame.
The research evaluates the effectiveness of shuffle-based Moving Target Defense (MTD) on host and network systems using a temporal graph-based security model, T-HARM. A novel dynamic security metric, the Reward-Based Metric (RBM), is introduced to assess the impact of MTD from the defender’s perspective, capturing changes in system resilience and attacker effort. The study involves implementing shuffle-based MTD techniques, defining and integrating the new metric with existing ones, and conducting simulation-based experiments to analyze security posture over time. The results show that the dynamic metric more accurately reflects real-time security changes, and that shuffle-based MTD significantly increases attack complexity and delays system compromise, thereby enhancing overall network defense.
The construction site layout is widely acknowledged as a complex challenge in the field of construction management. However, it is crucial for nearly all construction projects, as the layout of site facilities can significantly impact on the project costs. This issue can be formulated as a Quadratic Assignment Problem, which is widely recognized to be NP-hard problem. While numerous methods have been proposed, most existing approaches either lack sufficient global search capabilities or struggle to escape local optima, leading to suboptimal solutions. To address these limitations, this study proposes a hybrid GA-Tabu algorithm that integrates the exploration strength of Genetic Algorithms with the exploitation efficiency of Tabu Search. This integration aims to achieve a better balance between global and local search processes, thereby enhancing the overall solution quality for the CSLP problem. The proposed algorithm has been subsequently tested with three case studies previously utilized in related research, effectively demonstrating its performance and applicability.
This study investigates the patterns and determinants of Internet of Things (IoT) adoption among Malaysian small-to-medium enterprises (SMEs), examining organizational characteristics and implementation dynamics. Through comprehensive analysis, the research evaluates critical factors influencing IoT integration, encompassing external pressures, regulatory support mechanisms, technical capabilities, and organizational preparedness. The findings demonstrate that organizational age and size significantly influence IoT adoption propensity, with younger and smaller SMEs exhibiting greater inclination toward Industrial and Retail IoT implementations, primarily motivated by operational efficiency imperatives. While middle management emerges as crucial facilitators of IoT deployment, the study reveals significant disparities in workforce engagement and competency development. Implementation challenges persist across multiple dimensions, including financial constraints, technical complexity, and system compatibility issues. These findings underscore the necessity for targeted policy interventions, industry-specific capability development programs, and adaptable IoT solutions. The research highlights the strategic importance of leadership commitment, resource optimization, and ecosystem support in accelerating IoT adoption among Malaysian SMEs.
This study investigates the application of advanced Transformer-based models, namely BERT, DistilBERT, BERT-multilingual, ALBERT, and BERT-CNN, for sentiment analysis in Bahasa Malaysia, addressing unique challenges such as mixed-language usage and abbreviated expressions in social media text. Using the Malaya dataset to ensure linguistic diversity and domain coverage, the research incorporates robust preprocessing techniques, including synonym mapping and sentiment-aware tokenization, to enhance feature extraction. Through rigorous evaluation, BERT-CNN exhibits the best accuracy (96.3%), followed by BERT-multilingual (89.84%) and BERT (89.5%). DistilBERT and ALBERT delivered competitive performance (88.96% and 88.76%, respectively) while offering reduced computational requirements, highlighting the trade-offs between performance and efficiency. The study emphasizes optimized strategies for handling challenges in positive sentiment classification and demonstrates the efficacy of transformer architectures in nuanced sentiment detection for low-resource languages. These findings contribute to advancing Natural Language Processing (NLP) for scalable sentiment analysis across domains.
Internet of things (IoT) refers to devices connected to the internet that enable human-to-human or human-to-computer communication. It facilitates the collection and exchange of data between these devices and users. This study presents an IoT-based Smart Garden integrated with smart film technology, designed to predict light intensity and monitor daily plant growth. Despite a strong interest in gardening, many individuals lack the time to properly care for their plants. Therefore, this study proposes a simple and efficient method for plant monitoring and control using a smartphone or other connected devices. The system incorporates three primary sensors: the DHT11 sensor (for temperature and humidity), a soil moisture sensor, and a light intensity module sensor. The water pump and smart film function as actuators, which can be controlled remotely or via a smartphone. This IoT-based Smart Garden with smart film collects real-time data and transmits it to users via a mobile application, ensuring convenient remote monitoring and control. This research demonstrates that smart gardening systems significantly reduce human intervention, making plant care more efficient. Moreover, the sensors continuously gather and update environmental data, allowing users to stay informed about plant conditions in real time without needing to be physical present.
Efficient real-time object detection is a critical requirement in edge computing applications, such as smart surveillance, where resource constraints pose significant challenges. Existing deep learning methods often struggle to balance accuracy and efficiency, particularly when deployed on hardware with limited computational resources. This work focuses on developing a quantized object detection system utilizing advanced deep learning models to improve inference performance on edge devices, Zedboard and Jetson Nano. The Zedboard, an FPGA platform without GPU acceleration, executes a quantized YOLOv3-tiny model with ultra-low power consumption of 2.2W but requires over 3 seconds per inference, making it unsuitable for real-time applications. In contrast, the Jetson Nano, running an optimized YOLOv7-tiny model with FP16 quantization and GPU acceleration, achieves a processing speed of 38 FPS with mAP of 46.3%, while maintaining a low power consumption of 5.1W. Based on the results, this work presents a practical solution for real-time object detection in resource-constrained environments by demonstrating the benefits of combining quantized deep learning models with GPU acceleration. Future work could focus on fine-tuning models for specific applications, such as traffic monitoring, to improve the detection of vehicles, pedestrians, and traffic signs in dynamic environments.
Air quality, both indoors and outdoors, is crucial for public health as it affects respiratory conditions and overall well-being. Machine learning (ML) techniques offer innovative solutions for monitoring and predicting air quality. However, choosing the right ML algorithms and evaluation metrics is essential for creating accurate air quality prediction models, as these choices directly impact the accuracy and reliability of the results. This study aims to explore and summarize the literature on the use of ML techniques in predicting and optimizing air quality, addressing the urgent issue of air pollution. The review analyzed papers from two electronic databases, namely Scopus and Science Direct. Information on ML techniques for predicting air quality and identifying the main sources of pollution was extracted from 26 studies. The study focuses on common ML techniques employed in air quality prediction, including classification, deep learning, regression, ensemble learning, and combinations of regression and deep learning. It also identifies the evaluation metrics used to assess the performance of these models, such as recall, root mean square error (RMSE), mean absolute error (MAE), and coefficient of determination (R2). By synthesizing existing knowledge, this study provides a comprehensive understanding of the metrics researchers use to measure the overall effectiveness of ML algorithms. It serves as a benchmark for future research and guides the selection of appropriate evaluation metrics in the field of air quality prediction.
Adopting artificial intelligence (AI) and machine learning (ML) in software development processes presents an opportunity to systematically assess improvements in efficiency, accuracy, and project management. However, evaluating these technologies requires structured assessment models rather than generalized claims. This study utilizes a Kaggle dataset and applies linear regression, random forest classifiers, and K-means clustering to examine the impact of AI tools within Agile frameworks. The analysis reveals that AI tools enhance decision-making, productivity, and resource allocation in Agile environments. The linear regression model predicts willingness to adopt AI tools based on key variables, while the random forest classifier achieves high precision and recall in distinguishing AI tool users. Additionally, K-means clustering uncovers distinct adoption patterns among various roles, offering further insights into how AI adoption varies within Agile teams. Rather than assuming AI and ML’s impact, this study systematically evaluates their role in software development efficiency, providing a structured evaluation beneficial to both researchers and practitioners. While the findings highlight AI’s potential for optimizing Agile processes, they are constrained by the dataset’s scope. Future research should incorporate real-world industry validation and broader datasets to further substantiate AI’s effectiveness in Agile frameworks. This research contributes to the ongoing discourse on AI and ML adoption in software development, advocating for data-driven approaches in achieving scalable, efficient, and reliable software development processes.
