ABSTRACTWith the rapid growth of the Internet of Things (IoT) and the widespread deployment of smart connected devices, ensuring the security of these networks has become a critical challenge. Anomaly detection is considered one of the most effective techniques for identifying abnormal behaviors and cyber‐attacks in IoT networks. In recent years, deep learning techniques have gained significant attention in this domain due to their powerful capabilities in automatic feature extraction and modeling complex patterns. This review article provides a comprehensive overview of deep learning methods applied to anomaly detection in IoT networks. Various deep architectures including CNNs, LSTMs, autoencoders, GANs, and hybrid models are analyzed and compared. In addition, commonly used datasets such as CICIDS2017, BoT‐IoT, NSL‐KDD, and TON_IoT are introduced and evaluated in terms of their quality and suitability for deep learning‐based models. Key challenges including the lack of real‐world data, high resource consumption, vulnerability to adversarial attacks, and lack of interpretability are also discussed. Finally, potential future research directions are suggested to enhance the performance and real‐world applicability of deep learning‐based anomaly detection systems in IoT environments.

ABSTRACTModern power systems face increasing complexity due to fluctuating demand and the intermittent nature of renewable energy sources. To address these challenges, this paper introduces a novel Stackelberg game‐theoretic framework for intelligent demand response (DR) in smart grids. Our approach models hierarchical interaction between energy providers (leaders) and consumers (followers), incorporating priority‐aware load scheduling and real‐time feedback loops. Consumers are classified into priority and non‐priority categories. A Markov chain‐based behavior model captures stochastic user adaptation, enabling dynamic price adjustment. Simulations over 1, 7, and 30‐day horizons in MATLAB demonstrate significant improvements: A 22% reduction in operational costs and a 15% decrease in peak‐to‐average ratio (PAR). The framework converges efficiently and ensures adherence to the grid capacity. These findings demonstrate the effectiveness of our adaptive and scalable solution. Unlike existing Stackelberg‐based models, our approach uniquely integrates real‐time feedback, priority‐based user classification, and a stochastic Markov behavior model to enhance pricing responsiveness, grid reliability, and fairness across diverse consumer types.

ABSTRACTAiming at the problems of large computational complexity and high missed detection rate existing in the current printed circuit board assembly (PCBA) solder joint defect detection algorithms, and considering the requirements of rapid detection and easy deployment in industrial production, a PCBA solder joint defect detection method based on the lightweight improvement of the YOLOv8n model is proposed. Firstly, the backbone network is replaced with the HGNetV2 network. Through the effective extraction of multi‐scale features, the PCBA defect detection effect is improved and the storage requirements and computational resource consumption of the model are reduced. By replacing the Bottleneck module in C2f with the RepViTBlock module, a C2f‐RVB structure is designed, which improves the accuracy while reducing the number of parameters. The lightweight asymmetric detection head replacement further reduces model parameters and improves detection speed. Experimental results showed that the accuracy of the improved YOLOv8n model is increased by 5.9%, the number of parameters is reduced by 53.8%, GFLOPs are reduced by 53.1%, and model size is reduced by 49.2%. The experimental results indicate that the improved model can meet the requirements of real‐time detection of PCBA solder joint defects, and the relatively low number of parameters enables lightweight deployment of the model, which has certain reference value.

ABSTRACTEmployee turnover prediction remains a critical issue for organizations aiming to improve talent retention and minimize recruitment costs. The ability to predict when and why employees are likely to leave enables companies to take proactive measures to reduce turnover rates. This paper presents a systematic review of 58 studies focused on applying machine learning (ML) algorithms to predict employee turnover. We analyze various ML techniques, including Random Forest, Support Vector Machine, Logistic Regression, Decision Tree, and Neural Networks, highlighting their effectiveness in predicting turnover based on employee data. The review reveals that Random Forest emerged as the most widely used technique, achieving high predictive accuracy across multiple studies. Among the features, Job Satisfaction was identified as the most critical factor in turnover prediction, appearing in a majority of studies. Additionally, large datasets (more than 10,000 samples) were predominantly used, suggesting that more comprehensive data improve model performance. This review emphasizes the potential of ML in HR analytics and provides valuable insights into the strengths and limitations of each method.

ABSTRACT Agriculture faces critical challenges such as timely disease detection, fragmented market access, and limited use of real‐time technology in the field. To address these issues, we developed AgriSage, an Android‐based intelligent mobile application that integrates artificial intelligence, weather forecasts, and governmental scheme updates to support farmers, sellers, customers, and policymakers. The application incorporates two optimized deep learning models designed for on‐device deployment. The first model, based on MobileNetV2, performs binary classification to detect the presence of plants in images. It achieved a precision, recall, and F1‐score of 1.00 for both classes, indicating perfect classification performance on the test set. On‐device inference testing of the converted TensorFlow Lite model resulted in an average prediction time of approximately 3736.44 ms per image when evaluated through the validation pipeline. Another deep learning model, that is, a convolutional neural network designed for disease classification, was trained on the PlantVillage dataset across 38 classes. It achieved a macro average F1‐score of 0.8207 and a weighted average F1‐score of 0.8703. The optimized TensorFlow Lite version demonstrated an average inference time of 35.6 ms per image, confirming its suitability for real‐time, on‐device deployment. AgriSage delivers a robust and scalable platform integrating AI‐powered crop monitoring and disease detection. It also provides real‐time agricultural support services, contributing to improved decision‐making and promoting sustainable farming practices.

ABSTRACT The security of ultra‐low‐power Internet of Things (IoT) systems is critical yet challenging due to significant energy constraints. These networks are vulnerable to impersonation and data poisoning attacks, where malicious entities can mimic legitimate devices to gain access or corrupt system integrity. While traditional cryptographic solutions are often too energy‐intensive for these environments, radio frequency (RF) fingerprinting offers a promising physical layer security alternative by using intrinsic hardware imperfections to uniquely identify devices. However, existing RF fingerprinting methods often overlook the severe energy budgets of battery‐powered IoT devices. To address this challenge, this paper introduces two complementary deep learning models for device identification in long range wide area network systems. The first, RFNet, is a full‐capacity convolutional neural network that achieves 97.48% identification accuracy. The second, TinyRFNet, is an ultra‐lightweight model designed for resource‐constrained hardware, maintaining 93.19% accuracy with over 34 times fewer parameters than RFNet. We further propose a dynamic, energy‐aware inference strategy that adaptively selects between these two models based on the device's remaining battery level, the model's prediction confidence, and the operational context. Extensive experimental evaluation on a dataset of 30 commercial LoRa devices demonstrates that this adaptive approach achieves an overall identification accuracy of 94.54% while reducing energy consumption by 17% compared to exclusively using the high‐accuracy model. This system provides robust protection against physical‐layer threats with minimal energy overhead, thereby extending the operational lifetime of devices in secure, ultra‐low‐power IoT deployments.

ABSTRACT In the search for sustainable energy solutions, photovoltaic (PV) systems have emerged as a primary focus of innovation, attracting substantial worldwide interest in recent decades. Among the essential study topics, the development of control techniques, such as maximum power point tracking (MPPT), is a thriving and quickly growing discipline. Despite multiple advances, many control systems remain confined to simulation settings, far from practical deployment due to obstacles such as prohibitive cost (especially for researchers from developing countries) and system complexity. This disparity highlights the critical need for cost‐effective, dependable verification systems that can bridge the gap between theory and real‐world application. This paper therefore proposes the Real‐Climatic Microcontroller‐in‐the‐Loop (RCMIL) architecture, a pioneering platform for quick and realistic verification of MPPT controllers. By harnessing real‐climatic functionalities with Microcontroller‐in‐the‐Loop (MIL) execution, the RCMIL provides a practical‐like environment, allowing the implementation of PV control systems in real‐world situations, mimicking the nonlinear and stochastic character of weather patterns in any geographic region. Experimental results demonstrate the framework's efficacy, with the real‐climatic efficiency (RCE) reaching 97.15% under slowly changing conditions, while the real‐climatic mean absolute error (RCMAE) and real‐climatic mean absolute percentage error (RCMAPE) for voltage were measured at 4.55 V and 18.77%, respectively. Furthermore, the MIL feature allows the system to work on commercial microcontrollers, which reflects the real‐world problems that control algorithms would encounter in actual deployment circumstances. The study describes a clear, reproducible process for constructing the RCMIL platform and illustrates its efficacy through a series of experiments with proven MPPT controllers. The findings demonstrate the RCMIL's remarkable applicability for quick, low‐cost verification of PV control schemes, making it a valuable tool for researchers, engineers, and industry experts alike. By pushing the frontiers of PV system research, this paper provides a vital resource that speeds up the transition from simulation to real‐world applications.

ABSTRACTThis paper explores the intelligent integrated control and energy‐efficient technology applied to electric hydraulic excavators. Given that conventional control strategies struggle to achieve the desired balance of efficiency, energy conservation, and operational safety, an optimization approach based on intelligent integrated control is developed. The dynamic response characteristics of flow control are examined. This paper incorporates a fuzzy PID with a variable‐speed fixed‐displacement pump to design a control strategy. Subsequently, an electro‐hydraulic system model and a co‐simulation framework are constructed, followed by simulation analysis and real‐world validation. The findings indicate that the motor speed stabilizes within 1 s, demonstrating strong system stability and responsiveness. Compared to the fixed‐speed variable pump approach, the proposed method achieves a 44.3% reduction in total energy consumption (S8), aligning closely with the simulated result of 45.8%, with a minimal deviation of 3.3%. These results further confirm the effectiveness, rapid response, and stability of the energy‐saving control strategy within the intelligent integrated control system.

ABSTRACTRecently, deep eutectic solvents (DESs) have become known as solvents that have highly tunable properties and are eco‐friendly. Several terpenes, such as thymol and menthol, have resonance effects, which subsequently lead to their polar asymmetry, and this factor causes the majority of V‐type DESs to be investigated based on these terpenes. An External Electric Field (EEF) affects the intermolecular interactions and the electronic state energies of DESs. The ubiquitous presence of electric fields in systems plays an important role in functions. For this purpose, in the present study, MD simulations have been performed to design and describe the structural properties and gain deeper insights into the behavior of DES consisting of Menthol (MEN) and Thymol (THY) in molar ratios (1:1) in the absence (T = 298, 350, and 398.15 K) and presence of an EEF with strengths of 0.217, 0.434, 0.651, and 0.868 (V/Å) at a temperature of 398.15 K and atmospheric pressure. This is the first time that the structural and dynamical characteristics of DES‐based terpenes under EEFs have been investigated. The response of the eutectic mixture to the external fields is analyzed using the dipolar arrangements, the intermolecular hydrogen bond network, and the relative orientations of the nearest neighbors. The simulation results showed that as the external electric field increased from 0.217 to 0.651 (V/Å), the distribution of solvent species around each other increased. The results indicate that with increasing temperature, the hydrogen bond network between components HBA and HBD was disturbed at high temperatures.

ABSTRACT This study examines the effects of sonication duration, molding temperature, and weight percentage of MWCNTs and nano silica (SiO 2 ) on the inter‐laminar shear strength (ILSS) and Izod impact strength of laminate composites composed of glass fiber‐reinforced polymers and basalt. The laminates were made using the manual lay‐up approach and compression molding, with MWCNTs and SiO 2 added in equal amounts (0%, 1%, and 2% by weight). The ASTM D256 and D2344 criteria were adhered to while assessing mechanical properties. A total of 29 trials were made utilizing the Box–Behnken Design (BBD) of response surface technique, with the following independent variables: temperature, sonication duration, filler content, and molding pressure. According to an ANOVA analysis, these traits were significant for both ILSS and Izod impact strength. The ANOVA results also showed that the filler weight (A) is the most significant factor affecting the ILSS and Izod impact strength of hybrid nanocomposites, with molding temperature, pressure, and sonication duration coming in second and third, respectively. For design run orders three and eight, the ILSS values expressed were 24 MPa for the minimum and 40 MPa for the maximum. Izod impact strengths were 203 kJ/m 2 in design run order 3 and 167 kJ/m 2 in design run order 8. The optimal mechanical performance was determined by optimization using Design Expert 13 software at 2% filler content, 20 min of sonication, 5 MPa pressure, and 75°C molding temperature. This resulted in an impact strength of 201.47 kJ/m 2 and an ILSS of 40.25 MPa. When compared to the samples with the lowest performance, these findings show improvements of 40% and 18%, respectively. Additionally, the morphological features of cracked surfaces were revealed by scanning electron microscopy (SEM) examination, which shed light on the structural integrity of the composite.