Conventional misinformation detection approaches primarily rely on textual features and deep learning (DL) classifiers, which often fail to capture complex relationships among biomedical entities and the underlying scientific context of health claims. To address this limitation, this study proposes a graph neural network (GNN)-based biomedical misinformation detection framework that integrates knowledge graph propagation with semantic consistency verification. Initially, key biomedical entities such as diseases, treatments, and biological processes are extracted and mapped into a structured biomedical knowledge graph (BKG) to represent semantic relationships. A graph attention network (GAT) is then employed to model relational dependencies and propagate contextual information across connected entities, enabling the detection of hidden inconsistencies in biomedical claims. The proposed model is evaluated using benchmark biomedical misinformation datasets, including Reliable COVID-19 News Dataset, 2021 (ReCOVery), COVID-19 Healthcare Misinformation Dataset, 2020 (CoAID), and 2018–2020 biomedical health news corpus (HealthStory). Experimental results demonstrate that the proposed framework achieves an average detection accuracy of 96.3%, outperforming conventional long short-term memory (LSTM), convolutional neural networks (CNN), and transformer-based models in terms of precision, recall, and F1-score. The findings highlight that integrating structured biomedical knowledge with graph-based reasoning significantly enhances the reliability and interpretability of misinformation detection systems.
Existing deep learning approaches often exhibit limitations in contextual comprehension, high computational overhead, and restricted generalization when processing large-scale, tweet-level, and semantically ambiguous text. Moreover, deploying such computationally intensive models in real-time internet of things (IoT)-enabled monitoring systems and embedded platforms introduces additional constraints related to latency, memory footprint, and energy efficiency. To address these challenges, this work proposes a scalable hybrid deep learning framework (SHDLF). The proposed framework effectively captures semantic, syntactic, and temporal dependencies in both short and long social media texts through a novel integration of transformer-based representations and attention-driven feature fusion mechanisms. The architecture is designed with a modular and parallelizable structure to facilitate hardware-aware optimization and potential deployment on embedded and reconfigurable computing platforms, enabling efficient edge-level processing of high-velocity Twitter streams. Extensive experimental evaluations conducted on a large benchmark Twitter dataset demonstrate that SHDLF consistently outperforms state-of-the-art models, including convolutional neural network (CNN), bidirectional long short-term memory (BiLSTM), and baseline bidirectional encoder representations from transformers (BERT)-based architectures, in terms of accuracy, F1-score, and robustness under noisy conditions. The results confirm that SHDLF offers a robust, scalable, and computationally efficient solution for extracting reliable sentiment insights from noisy and dynamically evolving social media data.
This research presents an optimized multiple accumulate (MAC) unit multiplier design for efficient convolutional neural network (CNN) operations. This design mainly focuses on making the multiplier systems smaller by using approximate majority compressor methods instead of the usual and traditional approximate methods. The traditional approximate multiplier compressor techniques are leads to increases in logic size, critical path delay, and power consumption; however, the proposed research mitigates these problems and solves them with a novelty-based approach in the Dadda multiplier technique. The novelty of this approach is to reduce the number of stages in the multiplier design using 4:2, 5:2, and 7:2 compressors. This compressor is designed with an approximate method using majority logic; compared to this traditional method, the proposed majority approximate compressor method processed less error differences in multiplication output. The proposed approaches resulted in significant reductions in area, power, and delay relative to traditional multipliers. This research compared seven unique comparisons of MAC-based multiplier architecture, and it will have been developed in Verilog hardware description language (HDL) and synthesized on the Xilinx Vertex-5 FPGA, providing reductions of 58.4% in lookup table (LUT) and 76.2% in occupied slices, and proving less power consumption. This design is a highly suitable approach for real-time CNN and digital signal processing (DSP) applications.
Precise robotic systems often require multiple motor types, which increases hardware complexity, cost, and synchronization effort. This study presents an open-source multi-mode motor control platform based on four half-bridge power stages, enabling direct current (DC), brushless direct current (BLDC), and step per motor control on a single hardware architecture. Unlike existing software based multi-mode approaches, the proposed system introduces automatic motor type identification and safe connection verification at the hardware level, re quiring only a microcontroller and a integrated power stage. This represents a key novelty of the platform. Experimental validation was performed using three different motor types. The system achieved correct motor classification over re peated identification tests, with no false detections. Position control experiments confirmed stable operation across DC, BLDC, and stepper motor. The results demonstrate that the proposed platform significantly reduces system complexity while providing reliable multi-motor operation in a compact and low-cost structure.
The proposed compact wideband antenna is developed to meet the increasing demand for efficient and miniaturized radiators in sub-6 GHz fifth generation (5G) and internet of things (IoT) wireless systems. The design features an octagonal radiating patch integrated with modified H-shaped slots to enhance the current path and impedance matching, while a graded defected ground structure (DGS) is introduced to improve bandwidth (BW) and suppress unwanted surface wave effects. Fabricated on an FR4 substrate and energised by a simple stripline feed, the antenna maintains a compact size of 18×15 mm² without compromising performance. It achieves a wide fractional BW of 42.81% spanning 3.1–5.8 GHz, with a resonance centered at 4.6 GHz and obtained reflection coefficient of −36 dB, indicating excellent impedance matching. Additionally, the suggested antenna provides the maximum gain of 3.14 dB and an overall radiation efficiency of 80.5%, demonstrating stable radiation characteristics while making it ideal for small, low-profile 5G and IoT communication devices.
Existing elevator control systems in office buildings primarily rely on reactive scheduling strategies that respond only after passenger requests occur, leading to increased waiting times during peak traffic periods. Although reinforcement learning (RL) and deep learning approaches have been explored for intelligent elevator control, many existing methods require high computational complexity and large training datasets, limiting their suitability for embedded elevator controllers and practical smart-building deployment. To address this gap, this paper proposes a lightweight predictive elevator control framework based on the eXtreme gradient boosting (XGBoost) machine learning algorithm for rest-floor prediction. The proposed method uses historical traffic patterns and temporal features to predict future demand floors and proactively reposition idle elevators before passenger requests occur. A comprehensive simulation was conducted for multiple office-building configurations with varying numbers of floors and elevators over one year of operation using realistic traffic patterns. The proposed predictive strategy was compared with a conventional reactive control approach. Results show that the proposed framework reduces cumulative passenger waiting time by approximately 11%–22%, with larger improvements observed in high-rise and high-traffic scenarios, while maintaining comparable energy consumption. The study demonstrates that lightweight supervised machine learning can provide an effective and computationally efficient solution for predictive elevator control in embedded smart-building systems.
The rapid growth of online learning platforms has increased the need for intelligent systems capable of monitoring student attentiveness in real time to improve learning effectiveness and adaptive instruction. This paper proposes a multi-modal deep learning framework for attentiveness assessment by integrating visual, behavioral, and temporal information extracted from online classroom interactions. The proposed system consists of four major components, namely data acquisition, preprocessing and normalization, deep feature extraction with temporal learning, and attentiveness evaluation with analytics generation. Visual and spatial characteristics are learned using a convolutional neural network (CNN), while temporal behavioral patterns are captured through a long short-term memory (LSTM) network to model sequential engagement dynamics. The framework is designed to operate in both real-time and offline modes, enabling live monitoring during virtual classes as well as post-session analysis of recorded lectures. The computational pipeline is optimized through fixed-point processing, parallel convolution execution, and latency-aware temporal modeling, making it suitable for field programmable gate array (FPGA)-based and embedded implementations under constrained computational resources. Experimental evaluation conducted on an in-house dataset demonstrates that the proposed framework achieves 92.9% classification accuracy and a 91.9% F1-score, while maintaining strong generalization capability on cross-dataset benchmarks. Furthermore, latency analysis shows an average processing time of 31.6 ms per frame, enabling near real-time inference at approximately 30 frames per second.
Regarding dual industrial, scientific, and medical (ISM) band utilization, a planar, high-gain, dual-band modified antenna has been implemented application. This modified antenna features a rectangular patch with combined slots. This antenna has a profile of approximately 0.25 λ0×0.18 λ0. The combination of slots and modified rectangular patch allows for multiple-band performance. The designed antenna operates in two bands: 5.81 GHz and 2.42 GHz wireless body area network (WBAN). The antenna offered maximum radiation efficiencies of 76.4% and 82.8% in the two operating bands, with peak gains of 3.43 dB and 3.81 dB. The suggested antenna has reflection coefficients of -28.3 dB at 2.43 GHz and -23.9 dB at 5.81 GHz, respectively. The antenna's safety features were additionally evaluated employing a threelayer human body phantom initiated of fat, muscle, and skin tissues. The specific absorption rate (SAR) of the proposed antenna was evaluated using a three-layer human tissue model representing skin, fat, and muscle. The calculated SAR values were analysed according to the IEEE C95.1-1999 and IEEE C95.1-2005 safety guidelines, and the results confirm that the antenna operates within the permissible exposure limits. The measured results closely match simulations, demonstrating its reliability. Owing to its compact size, improved efficiency, strong impedance performance, and validated safety compliance, the proposed antenna is much impressed for effective ISM band communications.
The Matter protocol, created by the connectivity standards alliance (CSA), comes in with a single standard to make sure these devices can connect and be controlled across platforms like Google Home, Apple HomeKit, Amazon Alexa, and Samsung SmartThings. The rapid expansion of the internet of things (IoT) is driving the urgent need for secure and efficient onboarding processes for a wide range of connected devices. It necessitates a robust framework to seamlessly integrate new additions into existing systems while upholding security standards. This initiative focuses on implementing the Matter protocol on ESP32 devices, employing a Raspberry Pi hub as the central communication point to facilitate smooth device-to-hub interactions. This work presents the onboarding devices for interconnected IoT systems using the Matter protocol. The Matter device is configured and tested within the Amazon ecosystem using an Alexa Echo Dot, as well as with the smart home assistant ecosystem along with a smartphone application. By configuring the Raspberry Pi hub as a designated Matter hub and exploring interactions within the home assistant ecosystem supporting diverse platforms like Apple HomeKit and Google Home, the work enhanced interoperability and broadened the utility of IoT devices within an interconnected network. This initiative forges a foundation for an adaptable and cohesive IoT environment.
The rapid growth of digital learning platforms has generated large volumes of student interaction data, providing opportunities for intelligent prediction of academic outcomes. Beyond educational analytics, such prediction tasks are relevant for reconfigurable systems, embedded platforms, very large scale integration (VLSI) accelerators, and internet of things (IoT)-enabled edge devices in smart learning environments. This study proposes a hybrid machine learning framework for predicting student performance using the e-learning student reactions dataset, which captures engagement patterns, behavioral responses, and interaction dynamics. Eight classifiers— eXtreme gradient boosting (XGBoost), K-nearest neighbors (KNN), decision tree (DT), random forest (RF), support vector machine (SVM), multilayer perceptron (MLP), radial basis function (RBF), and deep neural network (DNN)—are evaluated using both an 80–20 train–test split and K-fold cross-validation to assess accuracy and generalization. Results show the RBF model achieves the highest accuracy of 1.00, demonstrating its ability to capture complex, nonlinear behavior. From a systems perspective, the framework can be mapped onto field programmable gate arrays (FPGAs) or embedded devices, leveraging parallel computation for low-latency inference, and integrated with IoT-enabled smart classrooms for real-time edge analytics. These findings confirm that hybrid machine learning models not only improve student performance prediction but also serve as practical workloads for reconfigurable, embedded, and VLSI-based intelligent systems in digital education.
Feature selection is critical for embedded machine learning systems where computational resources and memory are severely constrained. This paper presents the binary quadratically interpolated hybrid pathfinder algorithm (BQIHPFA), a novel metaheuristic optimization method designed for efficient feature subset selection in resource-limited classification tasks. BQIHPFA adapts the continuous QIHPFA to binary search spaces through sigmoid transfer functions and employs a hybrid two-group enhancement strategy combining pathfinder dynamics with salp swarm algorithm-inspired exploration. We evaluate BQIHPFA against three established binary optimization algorithms (binary particle swarm optimization (BPSO), binary grey wolf optimizer (BGWO), and binary whale optimization (BWO)) on three benchmark datasets with varying dimensionalities: Língua Brasileira de Sinais (Brazilian Sign Language) movement (90 features), Parkinson's disease detection (22 features), and Sonar Rock vs. Mine (60 features). Experimental results demonstrate that BQIHPFA achieves competitive classification accuracy (average 83.57%) with substantial feature reduction (average 64.1%) while executing 5.2 times faster than complex baselines and consuming minimal memory (peak: 45-58 MB). Ablation experiments demonstrate that every algorithmic part makes a 8-24% contribution to the total performance. BQIHPFA offers an easy-to-use, non-specific feature selection method to automated resource-constrained embedded classification systems, applicable to be deployed to low-power computing environments, and internet of things (IoT) edge systems.
Quadratic assignment problem (QAP) is one of the most difficult NPhard combinatorial optimization problems with applications ranging from facility layout design, scheduling and manufacturing systems to software optimization. This paper introduces a hybrid metaheuristic algorithm based on genetic algorithm (GA) and cuckoo search (CS); an improved solution quality and convergence speed is observed for QAP instances where GA itself performs poorly as well. This approach leverages the exploration capabilities of GA with computationally intensive exploitation that is easy for CS, to create a balanced yet robust searching mechanism across complex optimization landscapes. We tested the algorithm on benchmark instances taken from quadratic assignment problem library (QAPLIB) and compared it with many classical heuristics such as standard GA, particle swarm optimization (PSO) method and original CS algorithm. Experimental findings showcase that the presented hybrid GA-CS algorithm outperforms traditional standalone GAs regarding solution quality and computational time with significance by promptly converging toward high-quality solutions, especially for medium- to large-scale test instances. In addition, performance improvements over competing methods are shown as statistically significant using the Wilcoxon signed-rank test. The results show that the proposed hybrid framework is an efficient and accurate optimization technique for solving challenging QAP.
Biomedical signal processing is essential for modern diagnostics, monitoring, and preventive healthcare in public health and mobile health (mHealth) systems. Signals such as electroencephalography (EEG), electromyography (EMG), and heart rate variability (HRV) offer vital insights into brain, muscle, and cardiovascular health. However, achieving real-time, energy-efficient, and scalable processing remains challenging for conventional hardware such as central-processing units (CPUs), graphics-processing units (GPUs), and application-specific integrated circuits (ASICs). Field-programmable gate arrays (FPGAs) provide a promising alternative through their reconfigurability, parallelism, and adaptability to dynamic biomedical workloads. This review examines FPGA-based implementations for EEG, EMG, and HRV processing, focusing on key metrics including latency, throughput, and power efficiency. It also discusses design strategies such as low-power optimization, hardware–software co-design, and FPGA-based machine learning acceleration, with attention to data integrity and security in medical contexts. Integration with wearable, portable, and telemedicine platforms is explored, alongside comparative analyses with traditional computing architectures. The paper identifies challenges in power–performance trade-offs, design complexity, and clinical validation, and highlights emerging directions such as artificial intelligence (AI)-driven FPGA platforms, neuromorphic design, and sustainable low-cost solutions for large-scale health monitoring. Overall, FPGA-based biomedical signal processing emerges as a foundation for intelligent, efficient, and accessible next-generation public-health technologies.
Accurate object detection under low-light conditions is a critical requirement for reliable perception in autonomous driving systems. However, night-time environments often suffer from poor illumination, noise, and reduced feature visibility, which significantly degrade the performance of conventional object detection models. To address this challenge, this paper proposes spatial contrast learning (SCL)-you only look once version 11 (YOLOv11), an enhanced object detection framework designed for night-time scenarios. The proposed approach integrates SCL to improve feature discrimination in dark regions and employs the revolution optimization algorithm (ROA) for effective model parameter optimization. The framework is evaluated on three benchmark night-time datasets, ExDark, LLVIP, and BDD100K, to assess its detection performance. Experimental results demonstrate that the proposed model achieves a mAP@50 of 72.9%, improving the baseline YOLOv11 by 9.5% while also reducing inference latency by 18.3%. Comparative evaluations with existing detectors further confirm that the proposed method provides improved accuracy and efficiency for night-time object detection. These results indicate that the proposed framework can enhance perception reliability for autonomous driving applications operating in low-light environments.
This paper presents an artificial intelligence (AI)-assisted optimization framework for on-off current feedback controlled (ONOFIC)-enhanced domino circuits implemented in advanced fin field-effect transistor (FinFET) and carbon nanotube field-effect transistor (CNTFET) technologies. The framework integrates artificial neural network (ANN) surrogate modeling with evolutionary optimization (genetic algorithm (GA), particle swarm optimization (PSO), and NSGA-II) to reduce leakage, improve energy efficiency, and enhance robustness under process voltage temperature (PVT) variations, aging effects (bias temperature instability (BTI)/hot carrier injection (HCI)), and antenna-induced parasitic coupling. By replacing repeated HSPICE simulations with fast ANN predictions, the proposed methodology reduces computational cost by more than 90% while achieving up to 30–35% gains in leakage and power-delay product (PDP)/energy-delay product (EDP) performance. The results demonstrate that ANN-assisted evolutionary optimization provides a scalable and technology-agnostic workflow suitable for next-generation internet of thing (IoT), radio frequency (RF)-integrated, and low-power very large scale integration (VLSI) platforms.
Fires endanger not only the environment's wealth but also the entire fauna and flora, drastically disrupting a region's biodiversity and ecology. This article monitors the spread of a fire in a specific area, determines its approximate direction, attempts to extinguish it in an organized manner, and identifies the safest solutions. A new system has been developed, consisting of two parts: embedded and reconfigurable. This system comprises four accurate flame sensors, a buzzer, an Arduino, and a field-programmable gate array (FPGA) DEV board. The Arduino and FPGA collect data from these sensors and send it to MATLAB, which processes and displays the results. This paper also uses a two-stage prediction based on a spatial interpolation algorithm. The results showed that the speed and direction of fire spread could be predicted quickly and accurately using a spatial interpolation algorithm, achieving the lowest predictive error (mean absolute error (MAE) ≈0.33 and root mean square error (RMSE) ≈0.48) at approximately epoch 70. Moreover, the proposed method achieved a 92% success rate in detecting flames and fire flashes, indicating that the sensors respond to fires within under 1 minute of occurrence.
The advancement of photovoltaic (PV) systems in tropical regions faces significant efficiency challenges due to fluctuating panel surface temperatures. This study addresses these issues by implementing machine learning (ML) models, specifically k-nearest neighbors (KNN) and extreme gradient boosting (XGBoost), to classify and monitor panel temperatures. To enhance system resilience, an internet of things (IoT) based on-grid protection system was developed, featuring a dual-relay redundancy mechanism that triggers an automated trip when the current exceeds 1.30 A. This integration ensures the protection of both the PV infrastructure and household electrical loads. Experimental results demonstrate that the KNN model exhibits superior reliability with a testing accuracy of 93% and a baseline performance of 96.67%, successfully identifying both normal (25 °C to 35 °C) and high-temperature (36 °C to 48 °C) states. In contrast, while the XGBoost model reached a maximum validation accuracy of 94.44% during training, it only achieved a testing accuracy of 84% and showed significant limitations in detecting normal temperature patterns. Beyond classification, the IoT framework proved highly precise in real-time energy monitoring, with sensor error rates below 2%. This research offers a strategic solution for optimizing energy conversion and system reliability, providing a robust framework for sustainable clean energy management in tropical climates.
Driving cycles are speed-time profiles used to evaluate vehicle performance, fuel consumption, and exhaust emissions. However, real-world driving-cycle data for water vehicles are still limited, restricting accurate assessment of their energy efficiency and environmental impact. This study developed a low-cost water driving cycle (WDC) tracking device using an Arduino UNO integrated with global positioning system (GPS), global system for mobile communications (GSM), secure digital (SD) card storage, and an liquid crystal display (LCD) display. The device records speed, time, longitude, and latitude during water-vehicle operation. Prototype validation was performed by comparing the recorded speed with a standard GPS speedometer, while field testing was conducted along the Payang Water Taxi (PWT) route in Kuala Terengganu. The collected data were processed to construct a WDC and analysed using the advanced vehicle simulator (ADVISOR). Validation results showed percentage errors of 0.30% and 0.16%, indicating device accuracy within 5%. The ADVISOR analysis estimated fuel consumption of 24.1 L/100 km and emissions of 4.154 g/km HC, 2.851 g/km CO, and 0.08 g/km NOx. The proposed device provides a practical data-acquisition tool for water-vehicle performance evaluation.
The integration of sixth-generation (6G) communication and Industry 4.0 technologies has transformed industrial automation, connectivity, and intelligent data analysis. However, the increasing volume and diversity of data generated from multiple industrial sources create significant challenges for accurate and real-time fault detection. This study presents a deep learning-based framework designed to improve fault identification in 6G-enabled Industry 4.0 environments. The proposed system processes heterogeneous data collected from internet of things (IoT) devices, monitoring sensors, and automated industrial equipment to ensure reliable and scalable fault analysis. A hybrid model combining convolutional neural networks (CNNs) and long short-term memory (LSTM) networks is implemented to capture spatial features and temporal relationships within industrial datasets. The framework also focuses on optimizing computational resources while maintaining high detection performance. Simulation-based evaluations demonstrate that the proposed approach enhances fault detection accuracy and system reliability, making it suitable for advanced smart manufacturing and industrial monitoring applications.
Today, hybrid energy harvesters are critical in promoting technological advancement by generating sustainable energy and addressing the financial and environmental concerns around batteries. Because of their unexpected input behavior, hybrid energy harvesters present a challenge in producing the necessary stable energy. Thus, this study provides a power conditioning circuit with an optimal controller. Three proportional-integral-derivative (PID) controllers control the charging and discharging of the battery's bidirectional converter. To improve system performance actively and optimally, optimization algorithms are implemented for the optimization of the PID parameters. Osprey optimization algorithm (OOA)-based PID is used, and its performance is compared with five optimization algorithims (Chimp optimization algorithm (ChOA)-based PID, hony badger algorithm (HBA)-based PID, Zebra optimization algorithm (ZOA)-based PID, and cheetah optimization algorithm (COA)-based PID. The comparison between algorithms was done based on the minimum fitness function value, which shows that the OOA is the best one. All results are implemented in MATLAB/Simulink using the 2021a version as follows: (ChOA 3.061%, CO 4.737%, HBA 3.03%, ZOA 3.058%, and OOA 1.52%).
