Accurate and objective assessment of students’ knowledge remains a challenging problem due to the inherent uncertainty and subjectivity of traditional evaluation systems. Conventional grading approaches often fail to account for task complexity, discrimination power, and variability in student responses, which leads to inconsistent and biased results. This study proposes a multi-stage fuzzy logic–based decision-making model for knowledge assessment. The model integrates several key evaluation indicators, including task difficulty, discrimination index, response value, and response weight, within a unified fuzzy inference framework. A structured multi-factor evaluation mechanism is developed, where fuzzy membership functions and rule-based inference are used to transform qualitative judgments into quantitative assessment measures. Furthermore, a defuzzification process based on the Center of Gravity (COG) method is applied to obtain final scores, and a correction mechanism is introduced to refine evaluation outcomes. A comparative analysis was conducted using assessment data from 100 students across 5 tasks evaluated on a [0–10] scale. The results suggest that the proposed approach provides a more differentiated and consistent interpretation of student performance than the traditional assessment method. The proposed model provides a reliable and interpretable framework for evaluating students’ knowledge and supports the development of adaptive and intelligent educational assessment systems.

Vehicular Ad Hoc Networks enable dynamic and self-organizing communication among vehicles and roadside units, forming a fundamental backbone for advanced intelligent transportation systems. Efficient clustering plays a crucial role in VANETs by improving communication reliability, reducing network overhead, and enhancing scalability in highly dynamic environments. This study presents a comprehensive and critical survey of partitioning-based clustering algorithms in VANETs, explicitly addressing the lack of unified evaluation frameworks for distance metric selection and cluster quality assessment in dynamic vehicular environments. The significance of this work lies in its ability to bridge the gap between theoretical clustering approaches and their practical applicability in highly dynamic VANET scenarios through a structured and reproducible evaluation framework. Unlike existing surveys that primarily provide descriptive comparisons, this work introduces a structured and reproducible evaluation framework to systematically analyze the impact of distance metrics and clustering strategies under controlled simulation conditions. Widely adopted partitioning algorithms, including K-Means, K-Medoids, CLARA, and CLARANS, are systematically analyzed under diverse environmental conditions. Each algorithm is evaluated using multiple distance metrics, namely Euclidean, Manhattan, Minkowski, and Gaussian, to quantify similarity and dissimilarity among vehicles and to identify suitable clustering approaches for varying scenarios. The study identifies key research gaps, including the absence of standardized benchmarking, limited consideration of mobility-aware metrics, and insufficient analysis of distance metric sensitivity in highly dynamic scenarios. The quality of clustering is assessed using standard validation metrics, including Silhouette Score, Davies–Bouldin Index, and Calinski–Harabasz Index, along with cluster head lifetime to capture stability characteristics. Experimental results are presented as a supporting analytical component rather than a standalone contribution, with all simulation parameters, assumptions, and evaluation settings explicitly defined. The findings indicate that clustering performance is highly scenario-dependent, and while Euclidean distance and K-Means show strong performance under specific conditions, their effectiveness varies with network density, mobility patterns, and environmental dynamics. Overall, this study contributes to advancing the field by enabling more informed, reproducible, and context-aware clustering design, thereby supporting the development of more efficient and scalable intelligent transportation systems.

The rapid evolution of Internet of Things (IoT) networks has presented serious security threats because of the enormous volume of distributed data produced by connected devices. Traditional IDSs (IDS) usually follow centralized data collection, resulting in communication overhead, scalability issues, and privacy problems. Although federated learning (FL) offers a way to train distributed models while preserving privacy, many current FL-based AD techniques cannot be adapted to account for the interaction relationships between IoT devices. To address these challenges, this study introduces the federated graph attention network (FL-GAT) for anomaly detection in IoT-edge environments. The proposed framework treats IoT devices as graph nodes and introduces a multi-head graph attention mechanism to capture the spatial interaction among devices while guaranteeing data privacy by adopting federated learning. Local models are trained in a distributed manner on edge devices without sharing raw data. Distributed IoT attack scenarios were used to evaluate the proposed framework using the TON_IoT and Bot-IoT benchmark datasets. Experimental results show that FL-GAT achieved 95.2 % accuracy and 94.5 % F1 score on TON_IoT and 94.8 % accuracy and 94.1 % F1 score on Bot-IoT, with better results than centralized deep learning and federated deep learning baseline models, and graph-based baseline models. Furthermore, the attention mechanism enhances the interpretability of the model by identifying the key interactions between devices that lead to unusual activities. Although the proposed framework shows encouraging performance and scalability, the evaluation was conducted using benchmark IoT datasets under a simulated experimental setting. Future work will focus on real-world deployment scenarios, dynamic network conditions, and lightweight edge optimization for resource-constrained IoT devices.

Handover (HO) management in millimeter-wave (mmWave) fifth-generation (5G) networks faces critical challenges including limited propagation distance, blockage, and frequent disconnections, particularly in dense urban environments. Most existing solutions target high mobility scenarios, while dense urban traffic with low-speed heterogeneous environments and frequent stop-and-go scenarios remains under-explored. This study proposes a novel concept of cell pride where the neighbouring cells cooperate and select the best performing cell for each user equipment (UE) instead of competing with each other. Based on this idea, the Adaptive Cell Pride Traffic Load Balancing (ACPT-LB) framework is developed to enhance the reliability of handover and connection stability in 5G mmWave networks by combining cooperative cell selection, adaptive load balancing, and a neighbour discovery mechanism. The simulated results showed Handover Success Rate (HSR) of 100%, Ping-Pong Avoidance Rate (PPAR) of 100%, and Connection Stability (CS) of more than 88% for all simulations with UE densities ranging from 1000 to 5000, highlighting the effectiveness of the framework in low mobility and high-density urban environments. These results confirm that ACPT-LB offers a scalable and robust solution for mobility and traffic management in 5G and Beyond 5G networks.

This paper proposes an intelligent load balancing framework for distributed big data processing systems that integrates machine learning techniques with adaptive weight-based decision mechanisms. The study addresses limitations of traditional static load balancing methods, which do not account for dynamic workload variations and heterogeneous request characteristics, leading to inefficient resource utilization and bottlenecks in multi-node environments. The proposed approach combines an online learning model for real-time estimation of request complexity with multi-parameter evaluation of node states, including CPU utilization, memory consumption, queue length, response latency, and cache efficiency. A dynamic weighting strategy is used to construct an integrated load indicator for adaptive request distribution across nodes. The framework is deployed within a multi-layer distributed architecture consisting of clustered application servers, distributed databases, caching subsystems, and monitoring components, ensuring scalable and fault-tolerant processing. For evaluation, a three-node simulation environment was used with 10,000 heterogeneous requests, followed by extended testing on semi-realistic workload traces derived from web traffic patterns and database query logs. The dataset included over 1.2 million requests, capturing bursty arrivals, skewed distributions, and heterogeneous complexity. Experimental results show that the proposed method improves load distribution uniformity to 6%, reduces average response time to 210 ms, and increases throughput up to 13,800 requests per second. Statistical validation using confidence intervals and hypothesis testing confirms a 47% (±3.2% at 95% confidence level) reduction in mean response time and throughput improvement up to 14,200 requests per second under realistic workloads. 

The accelerating pace of digital transformation has expanded dependency on online services, exposing a widening misalignment between technology adoption and cybersecurity competence. This study investigates generational disparities in digital security literacy, perceived risk, and protective behaviors, with a particular focus on senior citizens as high-risk end users within computer network and information security ecosystems. 112 participants from various age and professional cohorts were surveyed using a four-point Likert scale with minimal central tendency response bias followed by both descriptive and mean-comparison analyses. Results demonstrate that cybersecurity literacy is medium but uneven (mean = 2.02), with respondents from the oldest age group (age sixty and above) reporting the lowest composite security score and weakest preventive practices, including a Two-Factor Authentication penetration of only 35%. There is a clear confidence-competence gap (+0.47) among senior citizens, meaning they tend to overestimate their ability to deal with digital services but underestimate the challenge of acquiring technical knowledge. Building on these findings, the paper introduces the Digital Guardian for Seniors framework a conceptual, human-centric intervention model integrating adaptive, visually oriented pedagogy, an intergenerational cyber-buddy system, and suggested metrics for longitudinal evaluation. The study contributes to computer network and information security research by providing demographic-based empirical evidence and outlining a theoretical foundation for future empirical testing and targeted interventions for an aging digital population.

This paper leverages the advantages of single-mode, high-bandwidth transmission in ridge waveguides to design a QV-band ridge waveguide 1-to-2 power divider and a four-port directional coupler. This addresses the issue of narrow single-mode operating bandwidth in traditional waveguide power divider-combiner structures, which is caused by internal multimode characteristics and electromagnetic discontinuities, thereby establishing an integrated power distribution and combining network; The power divider employs a ridge waveguide H-plane T-shaped structure to optimize impedance discontinuities and field distribution, while the radial combiner achieves efficient conversion from the TM₀₁₀ mode to the coaxial TEM mode through four-path radial ridge waveguide inputs and a central metal disk. Simulation results indicate that the ridge waveguide power divider has a relative bandwidth of 64% (31.1–60.39 GHz), while the radial combiner has a relative bandwidth of 31.4% (39.07–53.57 GHz). Using a back-to-back cascaded test setup, experimental verification was completed via a ridge-to-standard waveguide transition adapter. Within the 40–50 GHz operating band, the network exhibits a return loss greater than 15 dB and an insertion loss less than 1.2 dB, with excellent amplitude-frequency characteristics and phase consistency. This structure offers broadband performance, miniaturization, low loss, ease of fabrication, and potential for multi-channel expansion, providing a novel engineered solution for high-power microwave systems in the QV band.

Over the past seven years, significant changes have occurred in both the development paradigms and the practical use of software systems of varying complexity. These changes are largely driven by the rapid adoption of online artificial intelligence technologies based on large-scale language models. Such models are currently actively used in software development tasks, including source code generation and test plan creation, thereby integrating across various stages of the software development lifecycle. This article examines a classic research object—namely, the process of developing a system requirements specification—and proposes an approach to its formal verification using the ChatGPT online service. First, a detailed mathematical formalization of the research object is presented, followed by a structured model for preparing system requirements in projects using ChatGPT at various stages of development. Next, the proposed approach is illustrated using a real IT project example, demonstrating the sequential stages of requirements preparation in a modern development environment. The article defines the main categories of system requirements and discusses their representation in project documentation. To support the analysis, relevant tabular data and UML diagrams are provided. Furthermore, the study describes a methodology for formal requirements verification through prompt-based interaction with the ChatGPT system. The scientific novelty of this work lies in the application of requirements verification by modeling the expected behavior of the future software system using ChatGPT. Future research directions include incorporating a fifth category of requirements – business rules – using ChatGPT, which will enable modeling the behavior of the software system in real business processes.

This research aims to enhance predictive maintenance and inspection planning in urban construction projects. Recent advances in graph neural networks and graph transformer architecture have demonstrated significant potential for modeling complex lifecycle processes of building systems. However, most existing approaches remain predominantly data-driven and lack integration of physics-informed modeling and real-time data, which limits their applicability in large-scale urban environments. This research addresses this gap by proposing an approach for managing heterogeneous big data in urban construction projects, enabling prediction of the technical condition and inspection needs of structural elements. The core contribution is the development of a physics-informed heterogeneous graph transformer model that integrates domain-specific physical knowledge into the learning process through physics-based features and regularization mechanisms. The results confirm that all validation criteria are simultaneously satisfied: the difference between training and validation accuracy remains within the threshold (=0.05); The overall classification accuracy exceeds 92.06%; area under ROC curve above 0.8; F1-score is above 0.8 for all major classes; Physics-alignment error is lower than 0.15; and a strong Spearman correlation is observed between model predictions and physics-based indicators. The novelty of the proposed approach lies in the development of a physics-informed graph learning paradigm which enables the integration of structural mechanics, degradation processes, and heterogeneous data sources within a unified predictive framework.

Cloud computing has become an essential platform for business data storage and application management due to its scalability, accessibility, and cost-effectiveness. However, ensuring the security and privacy of sensitive cloud data remains a major challenge because cloud users do not have direct control over their stored information. Traditional single-biometric encryption approaches often suffer from issues such as biometric variability, spoofing risks, and single-point failure. To address these limitations, this paper proposes a multi-user multi-modal biometric encryption framework integrated with threshold-based access control for securing business data stored in cloud environments. In the proposed approach, fingerprint and audio biometric modalities are pre-processed to extract privacy-preserving feature vectors, which are fused to generate individual user-keys using a Hash-Based Key Derivation Function (HKDF). Subsequently, a master encryption key is generated through Shamir’s Secret Sharing and Lagrange interpolation mechanisms, where only a predefined threshold number of valid user-keys can reconstruct the master key for decryption. AES symmetric encryption is employed to secure the business data before cloud storage. Experimental evaluation was performed using the FVC2002 fingerprint dataset and Mozilla CV-Corpus audio dataset. The generated biometric master-key successfully passed Shannon Entropy, Chi-Square, Monte Carlo Pi, and decryption validation tests, demonstrating improved randomness, reliability, and resistance against unauthorized access. The proposed framework effectively eliminates single-point failure issues and enhances secure cloud data access through threshold-based biometric authentication.

 It is known that multicollinearity not only leads to the generation of redundant data as a result of data repetition, but also affects the stability of linear models of artificial intelligence and the reliability of results. The negative effects of multicollinearity can be seen especially clearly in the development of mathematical models of artificial intelligence algorithms. That is, the coefficients will be unstable in a mathematical model developed on the basis of a data set with multicollinearity. As a result of it, misconceptions arise in scientific conclusions drawn based on the coefficients. This article first discusses multicollinearity and its negative consequences in detail. In addition to, methods for determining multicollinearity in a data set based on the correlation coefficient, the variance inflation coefficient, and the condition index are discussed in detail. Moreover, this research paper analyzes the methods of eliminating multicollinearity by removing, combining features, and Principal Component Analysis. At the same time, the research will investigate the impact of multicollinearity on machine learning models such as LogisticRegression, LinearRegression, LinearSVC, and XGBClassifier using a multicollinearity dataset. The results of the study showed that eliminating multicollinearity leads to an increase in the accuracy of all considered artificial intelligence models. In particular, the ROC value increased by 0.102 in the Logistic Regression model, by 0.129 in the Ridge Classifier, and by 0.121 in the Linear SVC. Although the smallest difference value of 0.094 was achieved in the XGBoost model, the accuracy was higher than that of the other models. After the experimental results, the article presents conclusions and recommendations based on the results obtained.

The development of sixth-generation (6G) terahertz (THz) wireless systems requires equalization techniques that can effectively handle severe channel impairments while maintaining low computational complexity. In this work, we propose a hybrid equalization framework that fuses regularized zero-forcing (ZF) with maximum likelihood (ML) refinement for ultra-massive multiple-input multiple-output (UM-MIMO) systems. The proposed Regularized ZF and ML Fusion (RZF-ML) equalizer leverages a regularization factor to mitigate noise enhancement and ill-conditioned channel effects, followed by a lightweight ML-based candidate search that refines symbol detection. This design provides a trade-off between the simplicity of linear equalizers and the optimality of ML detection. Simulation results under Rayleigh and Rician fading channels with high-order quadrature amplitude modulation (QAM) demonstrate that the RZF-ML equalizer achieves significantly improved bit error rate (BER) performance compared to conventional ZF and minimum mean square error (MMSE) equalizers, while approaching ML detection accuracy at a fraction of its complexity. The findings suggest that the proposed method is a promising candidate for robust equalization in 6G THz UM-MIMO networks, enabling reliable high-capacity communication in challenging propagation environments.

Finding and managing malicious network protocols is still very difficult in cybersecurity due to sophisticated attacks and encrypted communications. This systematic review analyzes the 59 most recent studies from 2018 to 2025 discussing using Deep Learning to recognize malicious traffic. Importantly, the study proves that more people rely on transformer networks, consider self-supervised and blended approaches, and do not validate sophisticated systems in real time. In addition, it makes it clear that the data used, evaluation metrics, and methods for deploying models on hardware are not realistic enough. Quantitative synthesis reveals: CNN-based architectures dominate (42% of studies, mean accuracy = 96.8%), followed by hybrid CNN-LSTM models (22%, mean accuracy = 97.4%), while Transformer-based approaches (8% of studies) achieve the highest mean accuracy (98.2%) yet only 12% evaluate real-time latency; NSL-KDD remains the most frequent dataset (n=18, mean accuracy = 94.2%), whereas CICIDS2017 (n=14) yields higher performance (97.1% mean); only 6 of 59 studies (10.2%) report inference latency or throughput; and self-supervised or unsupervised methods appear in just 8.5% of studies despite demonstrating 96%+ zero-day detection capability. These statistically grounded findings provide a roadmap for developing deployable, real-time intrusion detection systems while exposing critical gaps in current research methodology.

Mobile devices have played a crucial role in enhancing education but students' concerns about security and privacy may act as a barrier to their engagement with mobile learning apps. We quantified how perceived security, privacy, risk and trust shape student adoption beyond TAM constructs. PRISMA-guided systematic review identified 34 studies from six databases. Random-effects meta-analysis pooled 28 correlations and a two-stage MASEM tested an integrated model. The results show that perceived security risk significantly diminishes student trust (β = -0.24) and the perceived usefulness of an app (β = -0.18). The trust strongly boosts both usefulness (β = 0.32) and positive attitudes (β = 0.29). The usefulness and attitude factors fully mediate the effect on a student's intention to use the app, explaining 79% of the variance (R² = 0.79). The trust is the linchpin for adoption. Security and privacy are not backend technicalities but frontend determinants that shape a student's initial decision to engage with mobile learning tools.

Wavelet analysis has established itself as a robust and highly effective framework for the processing and characterization of non-stationary signals. While classical dyadic wavelet transforms are widely utilized due to their computational efficiency, non-dyadic (rational) wavelet transforms often provide a superior representation of signal singularities and complex oscillatory patterns. The proliferation of diverse wavelet functions necessitates a systematic approach to basis selection, which remains a critical task for maximizing feature extraction capabilities.

This paper investigates fundamental approaches for evaluating the efficiency of wavelet bases, focusing on criteria derived from the energy distribution of decomposition coefficients, the similarity between the wavelet coefficients and the original signal, and mutual information metrics. The applicability and mathematical robustness of these evaluation methods are specifically examined in the context of non-dyadic wavelet transforms. To validate the investigated methodologies, an additive two-harmonic test signal is employed, subjected to four distinct types of interference (additive white Gaussian, impulse, pink, and multiplicative noise) under varying signal-to-noise ratios. Finally, a comprehensive Composite Quality Index (CQI) is proposed. By aggregating the considered energetic and information-entropic characteristics, this index provides a reliable criterion for selecting the optimal non-dyadic wavelet basis for specific signal processing tasks.

IoT networks face persistent security challenges due to limited compute, heterogeneous hardware, and weak threat-detection coverage. Classical machine-learning methods struggle with high-dimensional traffic and novel attack patterns. This paper proposes a hybrid framework combining Fractional Generalized Laguerre (FrGL) moment-based feature extraction with a Residual Network augmented by Squeeze-and-Excitation attention (ResNet-SE). FrGL moments yield compact, noise-resistant descriptors via simple recurrence relations, while ResNet-SE mitigates degradation in deep networks through identity shortcuts and adaptively recalibrates channels to highlight attack-relevant features. On the Bot-IoT and Leopard Mobile IoT benchmarks the method reaches 99.78 % accuracy and 99.37 % F1, exceeding KNN (84.7 %), MLR (87.5 %) and a baseline CNN (99.3 %); cross-dataset tests on UNSW-NB15 and IoT-Bot give 96.34 % and 97.12 % accuracy. The framework additionally delivers per-sample inference latency on server- and edge-class hardware (3.9 ms on an NVIDIA V100 and 27.4 ms on a Raspberry Pi 4B with a Coral USB accelerator), an energy cost of 0.42 J per inference on the edge platform, a sensitivity analysis over learning rate, batch size, fractional order λ and reduction ratio r, and an adversarial-robustness evaluation under FGSM and PGD attacks, supporting real-time deployment on resource-constrained IoT gateways.

With the extensive adoption of edge computing and IoT infrastructure, the vulnerability landscape has expanded significantly along with stringent constraints concerning computation, energy efficiency, and data privacy. Traditional centralized IDS solutions tend to be less than ideal for such conditions, as they are highly dependent on centralized data labeling, large-scale computation, and constant traffic sharing. This paper presents FedSSL-IDS, a novel privacy-preserving IDS framework leveraging Federated Learning (FL) and Self-Supervised Learning (SSL), specifically designed for the needs of edge-based network architectures. The solution applies autoencoder-based self-supervised learning to extract informative latent feature representations of unlabeled network traffic, after which federated learning is performed on the lightweight classifier with supervised learning without any raw data sharing. In order to facilitate the implementation of the system on resource-limited edge devices, the system employs advanced model optimization methods, such as magnitude-based pruning and post-training quantization. Performance evaluations of the FedSSL-IDS framework were conducted using the CICIDS2017 dataset in a simulated federated edge environment with class-imbalanced and non-IID client distributions. According to the experimental results, the full precision model reached an average detection accuracy of 96.90% across all classes, whereas the major attack classes, like DDoS and PortScan, achieved impressive class-wise accuracy rates. Moreover, the combination of pruning and FP16 quantization greatly decreases the size of the model and computational cost during inference without compromising its near-native accuracy in detecting intrusions. Nevertheless, aggressive INT8 quantization leads to a substantial reduction in the detection performance of rare classes of attacks such as SQL injection attacks, showing that a compromise must be made between efficient compression and reliable detection in edge scenarios. Even though the presented framework increases the privacy level since there is no raw traffic exchange in federated training, sophisticated privacy-preserving techniques like differential privacy and secure aggregation are not part of the current design.

obtained results demonstrate the significant potential of the proposed intelligent solution for extending computer vision capabilities in the monitoring of aviation objects and infrastructure. The experimental results also confirm the effectiveness of the OCR and fuzzy matching modules in improving object identification accuracy under real-world conditions.

Digital twins are revolutionizing various industries by enabling real-time monitoring, simulation, and optimization of physical entities through their virtual counterparts. However, the increased interconnectivity between the physical and digital realms introduces significant security and privacy challenges, necessitating the development of intelligent security models. This paper explores the architecture of digital twins and identifies the key characteristics of effective security solutions, such as adaptability, real-time response, data integrity, and privacy preservation. Through a comprehensive literature review, we highlight existing intelligent security frameworks that leverage machine learning and artificial intelligence technologies to address the growing range of cyber threats in digital twin environ-ments. Key observations indicate a trend toward integrating advanced analytics for threat detection and response, as well as the application of block chain technology to enhance data integrity and trust. Furthermore, this paper outlines future research directions, emphasizing the potential of innovations like federated learning, graph neural networks, and transfer learning to bolster security in digital twin systems. By examining these aspects, this work underscores the critical impor-tance of developing robust security frameworks to protect digital twins and ensure their safe deployment across various applications.

Tin-based perovskites are among the most promising candidates for high performance light-weight and radiation-tolerant space photovoltaics, but their response to energetic proton fluxes is not adequately determined. In this work, integrated SCAPS–SRIM analysis was applied to lead-free MASnI₃ perovskite solar cells for space applications in order to correlate device optimization with proton-radiation response. We established a combined SCAPS–SRIM simulation platform to simulate optoelectronic behaviors and radiation tolerance of an Au/Cu₂O/MASnI₃/TiO₂/FTO solar cell under AM0 illumination. Optimal-device calculations demonstrate that device absorber thickness of 0.20–0.30 µm and a TiO₂ ETL of 20–50 nm, Cu₂O HTL of 50 nm thicknesses result in good carrier collection and minimized recombination losses. Quantum efficiency and J–V measurement illustrate a stable operation under AM0 light, verifying the no extrinsic spectral incompatibility of MASnI₃ for the space energy source application. SRIM proton irradiation simulations (10-250 keV, 0° incidence) highlight the most damaging energy range within 50–150 keV for which masked Bragg peak lies in proximity to the MASnI₃ absorber and MASnI₃/TiO₂ interface accompanied by enhanced vacancy density, recoil energy deposition and phonon generation. High-energy protons (>200 keV) which deposit most of their damage in the rear contact stack, minimizing absorber degradation. The results overall indicate that MASnI₃ holds a good optoelectronic performance beyond the predictable radiation-damage behavior and thus can be considered as a promising alternative for space photovoltaic technology

This study introduces a unique framework that combines machine learning, and dosha profiling from Ayurveda, to improve precision, reliability, and interpretability of traditional diagnostic assessments. Four machine learning algorithms (Logistic Regression, Support Vector Machine (SVM), Random Forest, and XGBoost) were systematically investigated with a rigorous, quiz-based dataset that contained demographic, lifestyle, and physiological characteristics to classify six dosha categories (Vata, Pitta, Kapha, and their pairs). The experimental results showed a stark difference between linear and ensemble approaches. With an accuracy of 30% (F1 = 0.288), Logistic Regression provided marginal performance, suggesting there is limited separability in the overlapping patterns of health, while SVM came with an accuracy of 97.3% (F1 = 0.972) with kernel optimization. However, tree-based ensemble approaches improved predictive utility; Random Forest showed the highest overall performance (accuracy = 98.1%, F1-Macro = 0.982), with XGBoost and a Stacked Ensemble model behind (both ≈ 98%). This confirms ensemble approaches can represent the complex and nonlinear interdependencies associated with holistic wellness datasets. Interpretability analysis through feature importance ranking identified lifestyle and physiological variables—including sleep quality, appetite, emotional stability, skin texture, and digestion pattern—as the most important predictors, demonstrating a strong correlation to established Ayurvedic theory. Additionally, a desktop-based, interactive visualization was built to allow dosha prediction and wellness insights in real time. In conclusion, this work provides justification for Random Forest and XGBoost models as benchmarks for dosha classification and achieves a scientific syntheses of ancient Ayurvedic practice with contemporary machine learning. The results have important implications for the portfolio of digital Ayurvedic practice; to support data-informed personalized medicine; to foster new cross-disciplinary collaborations among ancient medicine, modern artificial intelligence and machine learning. This study compares Support Vector Machines (SVM), Random Forest (RF), and XGBoost for Ayurvedic dosha classification (Vata, Pitta, Kapha).

Wireless Sensor Networks play a vital role in the Internet of Things, smart cities, and industrial automation, yet there are open ended challenges in terms of efficient energy management and reliable data transmission. This paper presents a novel, two-phase routing framework comprising Dynamic Channel Selection and Energy-Efficient Routing Optimization to address these issues. In the first phase, Deep Q-Learning is utilized to identify stable communication channels, thereby enabling congestion-free data transfer across the network. The second phase implements Coral Reef Optimization to derive energy-efficient routing paths, significantly minimizing power consumption. Additionally, Adaptive Modulation and coding dynamically adjusts transmission parameters in real time to improve data throughput and reduce network delays. Existing solutions have been limited by network instability, poor scalability, and inefficient spectrum usage; In contrast, the integrated approach leverages Deep Q-Learning for intelligent channel allocation and Coral Reef Optimization for optimized route selection, while Adaptive Modulation and Coding fine-tunes the communication process to achieve optimal performance. Compared to existing models which shows high packet drop ratio and scalability constraints, our model achieves a 68% reduction in energy consumption, increases network lifetime by 82%, lowers error rate by 77%, enhances routing stability by 85%, and boosts overall throughput by 79%. These results highlight the proposed model’s potential as a highly adaptive, low-latency, and scalable solution for next-generation wireless sensor network applications.

The rapid rise of Large Language Models (LLMs) has shifted the battleground of digital misinformation. Unlike human-written fake news, machine-generated disinformation often employs subtle linguistic patterns that evade conventional detection systems. Although Deep Learning models can effectively identify synthetic text, they frequently operate as "black boxes," failing to offer the transparency needed for sensitive real-world applications. To address this, we introduce a hybrid architecture that merges the contextual strengths of DistilBERT with the sequential analysis capabilities of Bidirectional Long Short-Term Memory (BiLSTM) networks. Crucially, we incorporate SHapley Additive exPlanations (SHAP) to decode the model's decision-making process, visualizing exactly which words or tokens tip the scales toward a specific classification. Tests on the benchmark Fake or Real News dataset [1], supplemented by a 5-fold cross-validation protocol to ensure robust statistical validation, show our framework achieves an average accuracy of 96.92% ± 0.18%. By leveraging Explainable AI (XAI), we confirm that the model identifies actual semantic anomalies rather than merely overfitting to background noise, offering a more trustworthy foundation for automated fact-checking systems.

Unmanned Aerial Vehicles (UAVs) have become an effective solution for establishing emergency communication in post-disaster environments where conventional infrastructure is damaged. However, limited UAV battery capacity and unstable connectivity significantly reduce communication reliability and operational coverage. To address these challenges, this paper proposes an energy-efficient UAV-assisted communication framework based on Weighted Global Search Matrix Level (WGSML) clustering and optimal trajectory optimization for device-to-device (D2D) communication. The proposed WGSML method performs energy-aware cluster formation and cluster-head selection using residual energy, signal-to-noise ratio, and neighbourhood density. A Hidden Markov Model (HMM) is employed for routing optimization, while Q-learning-based resource allocation is utilized to determine optimal UAV trajectories and maximize residual energy utilization. Simulation results demonstrate that the proposed approach improves energy harvesting performance, reduces outage probability, minimizes computational runtime, and enhances spectral efficiency compared with existing clustering methods. The proposed framework provides reliable and sustainable communication support for post-disaster emergency response scenarios.

The rapid rise of the Internet of Things (IoT) has revolutionized connectivity across various domains, including smart homes, healthcare, and industrial systems. However, the large-scale integration of heterogeneous devices has significantly increased security vulnerabilities and cyberattack risks. Traditional intrusion detection systems (IDS) are often insufficient for IoT environments due to limited device resources and dynamic network behavior. This study proposes a machine learning–based IDS for detecting and classifying malicious activities in IoT networks in real time. Supervised learning algorithms, including Decision Tree, Random Forest, and Support Vector Machine (SVM), were employed to analyze network traffic and identify anomalies. Experimental evaluation using benchmark IoT datasets showed that the Random Forest model achieved the best performance with an accuracy of 98.1%, detection rate of 98.2%, precision of 98.0%, recall of 98.1%, and a low false positive rate of 1.9%. Comparative analysis demonstrated that the proposed approach outperformed conventional IDS techniques in both detection capability and reliability. These results highlight the effectiveness of intelligent learning models in enhancing IoT network security and supporting trustworthy network operations.