Software Defined Networking enables centralized control and dynamic programmability, but achieving efficient load balancing across distributed controllers remains challenging because of traffic variability and scalability constraints. Despite the significant potential of reinforcement learning for multicontroller load balancing, existing studies primarily focus on developing new learning architectures or switch migration mechanisms, with limited attention given to systematically comparing exploration strategies under identical network conditions. The experimental setup consists of twelve switches for data forwarding and three distributed controllers following east-west communication. This study compares five exploration strategies of Q-learning using throughput, latency, controller response time, load balancing efficiency, fairness, scalability, and congestion-related metrics. Existing studies also often neglect fault tolerance, energy efficiency, and real-world validation. Under congested conditions, SOFT Q-learning records the lowest throughput reduction of 0.92, followed by upper confidence bound with 1.56, epsilon-greedy strategy with 2.34, prioritized experience replay with 3.29, and Boltzmann exploration with 6.06. The epsilon-greedy strategy achieves the highest throughput of 7.88 and fairness of 0.9999, soft reinforcement learning records the lowest latency of 0.0115 seconds, upper confidence bound achieves the fastest controller response time of 0.0285 seconds, and Boltzmann exploration attains the highest load balance ratio of 0.9865.

5G IoT networks which use cognitive radio technology need dynamic spectrum access (DSA) to achieve fast response times during extreme environmental changes while managing extensive network operations. The research introduces a decentralized graph-based meta-reinforcement learning framework which enables cognitive IoT devices to learn spectrum access methods through decentralized learning. The proposed method uses Model-Agnostic Meta-Learning (MAML) with Graph Neural Networks (GNNs) to enable structure-aware few-shot adaptation which requires only local observations and neighbor interactions to function. Agents require between 1 and 5 gradient steps to develop new spectrum adaptation capabilities. The proposed framework achieved 83% spectrum utilization, an average throughput of 1.6 packets/slot, Jain's fairness index of 0.92, and a collision rate of 5%, outperforming Meta-RL and GCN-RL baselines. The study demonstrates that decentralized Meta-RL with relational learning offers an effective and scalable method for managing intelligent spectrum in future wireless IoT networks.

This paper presents a learning-augmented deterministic method for heterogeneous multi-UAV task allocation and routing in static, fully observed environments. UAV-specific asymmetric path-length matrices represent the output of an offline path planner, allowing online planning to focus on assignment and route construction. The deterministic core evaluates insertions, checks energy feasibility, prioritizes tasks with few feasible UAVs, repairs blocked states, and improves completed routes. A shallow multilayer perceptron ranks feasible candidates by predicting future regret, defined as the difference between the makespan after deterministic completion and the immediate post-insertion makespan. The objective minimizes makespan first and uses aggregate route length only when makespan values are equal within the prescribed numerical tolerance. The model was trained on 12 synthetic instances and evaluated on 20 separate instances, each containing 20 tasks and four heterogeneous UAVs under uniform, clustered, and mixed layouts. It was compared with two greedy insertion heuristics, the corresponding deterministic ablation (CT-D), and a linear-regret variant. The proposed method achieved a mean makespan of 106.24 versus 107.46 for CT-D, a descriptive reduction of approximately 1.1%. Compared with CT-D, aggregate route length increased by approximately 1.5%, while mean runtime increased from 0.059 to 0.233 s. Performance improved on clustered and mixed layouts but declined slightly on uniform layouts.

With the faster growth of fifth-generation (5G) and sixth-generation (6G) wireless communication systems, there are unprecedented requirements for ultra-low energy consumption wireless and microwave networks facing an increasing demand for enabling extreme-bandwidth data rates, massive connectivity, and low-latency communications. For optimization problems such as network routing, microwave resource allocation and antenna parameters adaptation, even though the general optimality of the solutions can be proven or demonstrated while optimizing directly with traditional algorithms, they often converge slowly and get trapped in local optima. In response to these issues, this paper presents an Energy Efficient Wireless and Microwave Network Framework using Advanced Hybrid Salp Swarm–Whale Optimization (HSSWO) algorithm. It provides a hybrid method that uses SSA with its good exploration ability and WOA for adaptively exploiting the routing paths, transmission power, microwave antenna parameters, and spectrum assignment. It further adds an AI-assisted network evaluation module for intelligent decision-making. The experimental results show that the proposed HSSWO framework attains 95.82% energy efficiency, a 41.6% improvement in network lifetime, throughput of 12.47 Gbps, packet delivery ratio of 99.21%, end–to–end latency of only 0.58 ms and a packet loss of only 0.69%. Moreover, the utilization of the proposed method leads to a reduction of optimization time for an average of 31.8% as well as improving convergence speed by an average of 36.4% over state-of-the-art optimization methods. The proposed HSSWO framework is an efficient energy-aware architecture for next-generation wireless and microwave communication systems as these results confirm.

The rapid expansion of memory and resource-constrained IoT devices has enormously increased vulnerability to cyber intrusions. Although deep learning-based intrusion detection systems (IDS) aim to improve intrusion precision, this improvement comes at the cost of increased inference latency and resource utilization. In this paper, we propose a lightweight CNN-GRU-based IDS model that utilizes mutual information-based feature selection to reduce input dimensionality, retaining the most informative features. The model is validated using four popular IoT security datasets: BoT-IoT, ToN-IoT, Edge-IIoTSeT and NSL-KDD. We employ SMOTE to reduce class imbalance in the training dataset for classifying both major and minor attacks. We achieve accuracies of 99.31%, 98.80%, 96.73%, and 94.20% on BoT-IoT, NSL-KDD, ToN-IoT and Edge-IIoTSeT respectively. Since inference latency is a critical requirement for resource-constrained IoT devices, the proposed model achieves inference times of 0.064 ms, 0.086 ms, 0.078 ms, and 0.073 ms on the respective datasets. These results demonstrate that the proposed IDS provides an effective balance between detection performance and computational efficiency for real-time IoT applications. In future we will focus on validating the proposed framework in real-world IoT deployment scenarios.

Anomaly detection in autonomous vehicle (AV) telemetry is a safety-critical task requiring accurate identification of abnormal vehicle behavior from multivariate kinematic sensor streams. This study presents a comprehensive comparative benchmark of twelve supervised classification algorithms and thirteen unsupervised clustering configurations applied to a real-world AV telemetry dataset comprising 112,028 observations and 23 engineered kinematic features after removal of seven constant-value columns. The dataset exhibits an approximately balanced class distribution (Normal: 56,380; Anomaly: 55,648). All classifiers were trained on an 80/20 stratified split and evaluated on Accuracy, Precision, Recall, F1-Score, and AUC-ROC. Clustering algorithms were evaluated on 15,000 randomly sampled observations using Silhouette Score, Davies-Bouldin Index (DBI), Calinski-Harabasz Index (CHI), Adjusted Rand Index (ARI), and Normalized Mutual Information (NMI). Results demonstrate that tree-based ensemble classifiers achieve performance approaching theoretical upper bounds — Extra Trees, Random Forest, Bagging, and Decision Tree (each F1 = 0.9998) — driven by the dominant discriminative power of speed, speed_time_ratio, rolling_distance_mean, and distance, which collectively account for 59.7% of Random Forest feature importance. In contrast, all unsupervised clustering algorithms fail to semantically recover ground-truth anomaly labels: the best-Silhouette configurations (Birch k=3, Silhouette = 0.9529) achieve ARI ≈ 0.000, while the best-ARI configuration (K-Means k=2, ARI = 0.440) recovers only partial anomaly structure. Three robustness experiments qualify these figures: 5-fold cross-validation confirms stability under random partitioning (F1 = 0.9998 ± 0.0001 for the top ensembles), while strictly temporal and unseen-scenario partitioning reduce peak F1 to 0.9750 and 0.9602 respectively, quantifying the contribution of window- and scenario-level information leakage to the near-saturated hold-out scores; a ten-sample sensitivity analysis shows the K-Means label alignment to be sample-dependent (ARI = 0.20 ± 0.21) whereas the degenerate high-Silhouette collapse is fully robust. Under the default configurations and static distribution assumption examined here, these findings indicate that supervised classification is the preferred paradigm for AV anomaly detection when labeled telemetry is available, and establish that Silhouette Score alone is a misleading model selection criterion in this context. All experiments were run with default hyperparameters in a documented software environment; robustness under hyperparameter tuning and distribution drift remains for future work.

The scheduling of tasks, usage of resources, and compliance with Service Level Agreement in dynamic and heterogeneous cloud systems are a challenge to cloud service providers. Traditional scheduling techniques are unlikely to manage the dynamism of workloads, resulting in the decline in performance, energy wastefulness and breach of Service Level Agreement. In this paper, the Broker-Driven Hybrid Wild Goose-Owl Optimization - Deep Reinforcement Learning Framework is suggested to integrate the Service Level Agreement-aware filtering of brokers with a two-level optimization pipeline. The broker filters the incoming tasks against SLA constraints. A Deep Reinforcement Learning agent makes the first task assignment depending on the system condition and projected SLA risk. The assignments are further optimized using a Hybrid Wild Goose-Owl Optimization algorithm to minimize the makespan, energy use, imbalance between the CPU processors, migration cost, and SLA violation rate. It has been experimentally demonstrated that the hybrid structure achieves a 32, 18, 58, and 74% reduction in the makespan, energy usage, CPU imbalance, and SLA violations, respectively, relative to baseline heuristics, and 41% reduction in SLA violations relative to DRL-only scheduling. These results prove that a combination of SLA intelligence broker and hybrid evolutionary and learning-based optimization can facilitate the management of cloud resources in a robust and scalable way.

Real-time mobile computing environments, such as smart cities and disaster-response networks, generate heterogeneous and high-velocity data streams in highly dynamic settings. Traditional full replication incurs prohibitive bandwidth overhead, while naive partial replication risks missing life-critical alerts. This paper presents Semantic-Aware Partial Replication (SAPR), a unified and lightweight framework that resolves this critical bandwidth-reliability trade-off by intelligently selecting the most valuable information for dissemination. SAPR is an integrated solution to combine RDF/SPARQL-based semantic ranking, real-time priority handling, and efficiency-per-byte optimization with Deep Q-Network (DQN) adaptive control. The framework computes a semantic priority score based on data type, urgency, and spatial context, and then replicates the most utility-efficient triples per byte. Crucially, a DQN agent dynamically adapts the replication factor (k) by observing network conditions like mobility dynamics, neighbor count, and alert frequency, learning to prioritize bandwidth conservation in congested environments. Using a 26-node mobile network simulation validated with real CRAWDAD taxi traces and a 10,000-triple RDF dataset, SAPR achieves an exceptional balance. The proposed RL-SAPR scheme demonstrates 82% bandwidth reduction while successfully maintaining 99% alert delivery accuracy, significantly outperforming both semantic and priority-only baselines. This work confirms the feasibility of combining semantic intelligence with reinforcement learning to enable reliable, bandwidth-efficient knowledge sharing in mission-critical, bandwidth-constrained mobile networks.

A mobile agent is a small piece of program that migrates automatically among different platforms, on which it executes assigned tasks. Mobile agents migrate in malicious and unsecure networks. So, there will be chances of compromising mobile agent as well as host computer. So, during the mobile agent executing life cycle there will be chances of stealing of confidential information about host and agents. The application of mobile agents is increasing day by day in various domains such as distributed computing, cloud computing and Internet of Things. In this article, we propose a method based on lattice-based dynamic threshold Chinese Remainder Theorem for the security of mobile agents. The combination of dynamic threshold, lattice-based encryption-decryption and Chinese remainder theorem provides optimal security as compared to traditional approach. After the analysis of turnaround time of creation of share, distribution and recreation of secret is optimal as compared to traditional approach. Another comparison is also done on the basis of memory utilization, computational cost, communication overhead and scalability of proposed method. Proposed approaches provide effectiveness of security for mobile agents in a decentralized and malicious environment. The framework was implemented in Python and analyzed with 10 to 1000 mobile agents. With 1000 mobile agents, the framework gains a turnaround time of 12.4 ms, memory utilization of 18.7 MB, and CPU efficiency of 91.3%, demonstrating better scalability and computational cost compared with Shamir Secret Sharing, Integer CRT, Homomorphic Encryption, Vectorized Tree Parity Machine and Boneh Goh Nissim techniques. Performance evaluation based on computational cost, communication overhead and scalability further shows the effectiveness of the framework. These outcomes show that the proposed methods provide an efficient, scalable and post-quantum-secure method for securing mobile agents in cloud computing, Internet of Things (IoT), and other decentralized distributed computing environments.

The growing sophistication of contemporary network infrastructures has increased the pressure on the smart and dynamic intrusion detection systems that can react to the dynamic cyber threats. The machine learning (ML) and deep learning (DL) methods have high classification rates, but they work with fixed decision boundaries which reduces their ability to adapt to dynamic traffic distributions and emerging attack patterns. In order to overcome these issues, the resource-aware Proximal Policy Optimization (PPO)-based adaptive multi-class intrusion detection system (IDS) is suggested in this study. The system characterizes intrusion detection as a sequential decision-making and incorporates computational resource measures into the reinforcement learning (RL) rewarding framework, which allows optimizing detection performance and operational efficiency at the same time. In the ensemble comparison, PPO-Model achieved the highest accuracy (99.4%), recall (98.6%), and Macro AUC (0.998), while reducing CPU utilization by 25.8% and memory consumption by 27.1% compared with the Stacking model. These results demonstrate that the proposed approach can improve detection performance while reducing computational resource requirements. The results suggest that next-generation intrusion detection in the dynamic network environment can be achieved with a scalable and robust solution based on the combination of RL and resource-aware optimization.

Signature-based IDS is becoming more difficult due to the ever-increasing amount of encrypted network traffic. Hence, the demand for interpretable and lightweight intrusion detection techniques becomes essential. In this paper, we propose Graph-Temporal Adaptive Intrusion Detection-Non-Parametric (GTAID-NP), which is a novel graph-temporal framework that exploits degree-based graph structure, temporal periodicity extracted using Fast Fourier Transform, and non-parametric log-odds estimation. We evaluate our proposed model on the BCCC-DarkNet-2025 benchmark dataset that contains 22,795 flow records, among which 6,317 are encrypted and 16,478 are not encrypted, which are expressed by 427 features. An extensive ablation study on 192 experiment setups was done to analyze the influence of bin granularity, total-variation smoothing, feature selection strategy, top-K feature selection threshold, graph augmentation, temporal periodicity extraction, and leakage-guard correction based on an 80:20 stratified train-test split. Among all experimental setups, the best setup achieves an AUC of 0.878. The results suggest that fine granular binning, weak smoothing, and joint consideration of graph-structure and temporal properties contribute to better detection performance. We also conduct robustness evaluation of GTAID-NP through five different random seed runs, which achieve a mean AUC of 0.9074 ± 0.0036, showing the stable performance of GTAID-NP on various train/test splits. While the ensemble methods outperform our model on the benchmark score, GTAID-NP can be regarded as a transparent and lightweight approach for intrusion detection on encrypted traffic. Future research works include adversarial robustness, federated inference, and adaptive online learning.

Proper land-cover and deforestation classification is crucial in environmental protection, sustainable land use, and climate change. Nevertheless, the current deep learning and remote sensing models are usually limited by poor spectral-spatial feature integration, high computational complexity, and poor feature selection, which reduces the classification accuracy. To address these limitations, this paper presents a new model that combines SpectroRes-Former, a SpectralSpatial Residual Transformer Network, and QuFiSel (QuantumFisher Selection), a hybrid feature selection method. The SpectroRes-Former effectively learns both spatial and spectral dependencies with a ResNet encoder, spectral-spatial attention fusion, and Transformer-based contextual learning. At the same time, QuFiSel improves the discrimination of features by integrating Quantum-Inspired Feature Selection (QIFS) and Fisher Score, so that the most informative features are selected to classify. The effectiveness of the approach is tested based on the Indian Pines hyperspectral image, which consists of 220 bands after typical band selection and 16 land cover types. The performance is measured through 5-fold cross-validation and the statistical validation through mean, standard deviation, and paired tests. The obtained average classification accuracy of the proposed method equals 99.21 ± 0.08%, along with the Precision value of 99.08%, Recall 99.16%, F1-score 99.12%, and Matthews Correlation Coefficient 0.991, demonstrating superior results compared to some of the most competitive state-of-the-art techniques. The paired t-test gives p < 0.05, which confirms the statistical significance of the result. It is shown that the introduced SpectroRes-Former framework can be used effectively for hyperspectral land cover classification while providing high generalization. The offered solution can serve as a reliable base for smart environmental monitoring systems.

The rapid expansion of IoT networks is leaving them susceptible to threats like DDoS attacks, which have the potential to affect the operation of vital services. This work proposes a hybrid intrusion detection system combining GPUaccelerated CuDNNLSTM and CNNLSTM models to capture both spatial and temporal traffic features. Using the Kitsune dataset with nine attack scenarios, the models were trained and tested in a Kaggle GPU environment (NVIDIA Tesla T4/P100, CUDA 11.x, CUDNN 8.x). The hybrid approach achieved over 98% accuracy, 97% precision, and ROCAUC above 0.98, outperforming classical ML baselines such as SVM and Random Forest. SHAP explanations provided transparency by highlighting key features behind each detection, while blockchain logging ensured tamperproof records of attack events. This system grants its users a clear view of the process logic by breaking it down into SHAP-based Explainable AI, which demonstrates decisive features for each decision. The attacks identified are stored in a safe manner on the Ethereum blockchain via smart contracts, making the solution difficult to tamper with using any intrusion technique. Therefore, the proposed approach presents a very viable option for reliable intrusion detection in an efficient and faster version for designing a DDoS detection system with new network configurations. Challenges include blockchain latency and deploying resource-constrained IoT devices. Future work will explore lightweight variants and federated learning to improve scalability.

The Bit Error Rate (BER) of Differential Chaos Shift Keying (DCSK) is analysed for both Single Input Single Output (SISO) and 2×2 Multiple Input Multiple Output (MIMO) configurations over multipath fading channels. Chaotic carriers are generated using a logistic map with control parameter r = 3.99 and spreading factor β = 128. Closed form BER expressions are derived for both architectures and cross validated through MATLAB simulations. In the SISO case, the derivation proceeds under Additive White Gaussian Noise (AWGN), while the MIMO model incorporates multipath fading and applies Equal Gain Combining (EGC) at the receiver to exploit spatial diversity. The 2×2 MIMO–EGC scheme reduces BER by approximately one order of magnitude relative to SISO at comparable SNR. Theoretical and simulation results agree well at low SNR; a divergence at high SNR is attributed to the Line of Sight assumption embedded in the analytical model, which sets the second-path coefficient to zero and therefore underestimates the diversity gain seen in simulation.

Federated Learning (FL) has become a promising distributed machine learning paradigm that allows collaborative model training while maintaining user privacy by keeping sensitive data on local devices. However, existing FL-based next-word prediction systems mainly focus on the model performance and lack robust mechanisms to prevent unauthorized users and compromised devices from joining the training process, raising security and reliability issues. To overcome this limitation, in this paper, a secure homogeneous federated learning framework for next-word prediction is proposed by combining a Long Short-Term Memory (LSTM) model with a dual-factor authentication mechanism. The proposed framework includes a central aggregation server and three homogeneous client devices with the same model architecture and training configurations to ensure stable convergence and consistent learning. The dual-factor authentication mechanism integrates the OTP-based user authentication and device authentication to guarantee that only legitimate users and trusted devices can join in the collaborative training. The main metrics for experimental evaluation were prediction accuracy, convergence speed, and security performance. The proposed framework achieved prediction accuracies of ~100%, ~100%, and ~98% across the three participating clients after 50 training epochs, showing faster convergence and more stable learning than a conventional federated learning baseline. Moreover, the authentication mechanism successfully resists unauthorized access with low computational and communication overhead. The results demonstrate that the proposed framework not only improves the accuracy and security of federated next-word prediction but also enhances the trustworthiness, reliability, and practical deployment of privacy-preserving language prediction systems in distributed edge environments.

Nowadays, in cloud computing, task scheduling is a key factor to enhance the performance of the distributed computing environment. As workloads vary dynamically, traditional scheduling algorithms like RR and HRRN have limitations in achieving optimal performance. To overcome these constraints, this paper proposes a PSO-Optimized Hybrid Scheduler that integrates Particle Swarm Optimization (PSO) with RR, HRRN, and Shortest Remaining Time First (SRTF) scheduling approaches. In this integrated approach, the PSO role is to find the optimal task execution order, the RR role is to ensure fair time-sharing among tasks, and the HRRN role is to prioritize processes to dynamically reduce excessive waiting time and prevent starvation. The SRTF provides short-job bias through the inverse remaining time term. The proposed mechanism is evaluated by using AWT and ATAT as key metrics. A Python-based simulation and a data set of six processes are used for validation and implementation, and a mathematical analysis is also presented to ensure correctness. The proposed PSO-optimized Hybrid Scheduler shows lower waiting and turnaround times than classical RR and HRRN scheduling approaches. In comparison with HRRN, the improvements of the proposed mechanism in AWT, ATAT, ART, and AS are 3.6%, 2.3%, 3.6%, and 7.9%, respectively. The findings are that the proposed mechanism provides an effective, balanced solution for resource scheduling in cloud environments.

Information about the mineral makeup and whether the surface has water-related compounds can be obtained from the lunar surface’s spectral reflectance. The Chandrayaan-2 Imaging Infrared Spectrometer (IIRS) and Chandrayaan 1 Moon Mineralogy Mapper (M³) are hyperspectral instruments that can capture detailed spectral signatures across a wide range of wavelengths, from visible to infrared. This lets scientists’ study lunar materials in great detail. This paper presents a systematic methodology for analyzing and interpreting the hyperspectral data from Indian lunar missions to derive significant reflectance data. The workflow includes converting measured radiance to reflectance, applying photometric adjustments, removing thermal radiation contributions using the Planck-based model, and getting spectra from certain pixels in selected lunar areas. To study two sites, IIRS images were taken from the following two craters: Gardner crater and Glauber crater. Glauber Crater’s images facilitated a direct comparison between the two missions. From the spectrum produced, the presence of iron-based lunar rocks like olivine and pyroxene can be identified. The absorption dip observed in Gardner crater within the 3 µm band proves the presence of hydroxyl or water on the surface. On the other hand, Glauber Crater lacks an absorption dip, indicating it is dry. The spectra obtained from IIRS and data are similar in spectral shape over the same area, indicating that both instruments independently confirm the absence of hydration for characterizing the lunar surface composition. With these findings, we gain insight into the Moon’s mineral content and composition, variations in mineral properties from one area to another, and the presence of water signatures that depend on topography. The thermally corrected IIRS spectrum of Gardner crater reveals a characteristic absorption feature at 3 µm centred at 2.918 µm with band depth of 0.072 and FWHM of 0.194 µm, suggesting that there is surface OH/O on Gardner crater as has been detected before by IIRS hydration features at similar latitudes of the Moon. On the other hand, the spectra from both IIRS and M³ at Glauber crater reveal only an increase in the slope of the spectrum without showing any 3 µm hydration absorption feature. The IIRS reflectance ranges from 0.025 to 0.200 while the M³ reflectance ranges from 0.015 to 0.038. This is expected due to variations in calibration of the two sensors; however, in this case it shows consistency in the fact that there is no surface hydration.

Energy preservation remains a crucial challenge in the Heterogeneous Wireless Sensor Networks (HWSNs) deployment. The uneven distribution of primary energy leads to precocious node depletion and network segmentation. Five intelligent routing and clustering optimization strategies tailored for dynamic and heterogeneous environments are presented and comprehensively compared in this paper: the proposed Binary-Chromosome Genetic Algorithm (BC-GA), Grey Wolf Optimization (GWO), Deep Q-Learning (DQL), Distributed Energy-Efficient Clustering (DEEC) and Power-Efficient Gathering in Sensor Information Systems (PEGASIS). The simulations are performed on a 100×100 m² field with 50 randomly deployed nodes having different initial energy levels (0.507- 0.986 J) communicating with a Base Station. Empirical results obtained from five trial Monte Carlo simulations expose excellent performance trade-offs. BC-GA achieves an outstanding Last Node Dead (LND) of 5339.60 ± 157.57 rounds and maximum throughput of 166,219.40 ± 2534.72 packets. Among the classical baselines, PEGASIS achieves an LND of 3095.20 ± 212.68 rounds and throughput of 120,742.00. On the other hand, GWO and DQL have accelerated depletion phases with FND 432.40 ± 62.15 and 568.80 ± 48.30 rounds, respectively, due to extreme pressure on the bottleneck node near the Base Station. The End-to-End (E2E) delay provided by DEEC is the minimum (11.49 ± 0.36 ms) among all methods and the Energy-Delay Product (EDP) is the best (6.09 ± 0.38) which makes DEEC very suitable for latency-critical environments. Those results prove that the evolutionary cluster-head structuring provided top-level load balance and an extension of lifetime for intensive data monitoring.

Wireless sensor networks enable real-time monitoring in critical applications such as industrial automation, environmental sensing, healthcare, and infrastructure management. However, sensor node faults caused by hardware degradation, communication failures, and harsh operating conditions can significantly reduce data reliability and system performance. Existing anomaly detection approaches often depend on large amounts of labelled fault data and suffer from class imbalance, noise sensitivity, computational complexity, and limited generalization in dynamic environments, highlighting the need for adaptive and scalable fault diagnosis frameworks. To address these limitations, this study proposes a self-supervised deep anomaly detection framework for wireless sensor network node fault diagnosis. The proposed method integrates preprocessing and sliding-window segmentation with a Deep Dual Masked Transformer (DDMT) architecture to capture temporal dependencies and learn discriminative features without extensive labelled data. A self-supervised reconstruction objective with a 15% masking strategy is employed to optimize latent representations, while an adaptive anomaly-scoring mechanism generates reliable fault predictions. Experimental results on the Zidi wireless sensor network dataset demonstrate that the proposed method achieves an accuracy of 99.0%, a precision of 98.5%, a recall of 98.3%, an F1-score of 98.4%, and an ROC-AUC of 99.5%, outperforming existing baseline methods. By combining temporal modelling and representation learning within a unified framework, the proposed approach improves anomaly detection accuracy, robustness, scalability, and generalization capability while maintaining low false alarm rates in challenging wireless sensor network environments.

The rapid evolution of next-generation wireless communication systems has increased the demand for intelligent microwave components and energy-efficient wireless architectures capable of supporting ultra-high data rates, low latency, massive device connectivity, and adaptive resource management. Conventional optimization approaches often experience slow convergence, high computational complexity, and limited adaptability when optimizing multiple communication parameters simultaneously. To address these challenges, this paper proposes a Machine Learning-Assisted Adaptive Pelican Optimization (ML-APO) framework for the joint optimization of microwave components and wireless communication architectures. The proposed framework integrates an Adaptive Pelican Optimization Algorithm with a machine learning-assisted surrogate prediction model that rapidly estimates communication performance metrics, thereby significantly reducing computational overhead during the optimization process. The optimization simultaneously considers transmission power, antenna parameters, microwave component characteristics, spectrum allocation, routing efficiency, and communication reliability through a multi-objective fitness function. The adaptive search strategy of the Pelican Optimization Algorithm effectively balances global exploration and local exploitation, enabling faster convergence toward optimal network configurations while avoiding premature convergence. Extensive simulation results demonstrate that the proposed ML-APO framework achieves 96.41% energy efficiency, 43.8% improvement in network lifetime, 13.56 Gbps throughput, 99.43% packet delivery ratio, 0.49 ms end-to-end latency, and 0.54% packet loss, while reducing optimization time by 35.8% and improving convergence speed by 39.6% compared with existing optimization methods. These results demonstrate that the proposed framework provides an efficient, scalable, and intelligent solution for designing high-performance microwave components and next-generation wireless communication architectures suitable for future 6G and beyond wireless networks.

This research aims to understand the impact of the network area and shape on the performance metrics used in evaluating the random walk-based source location privacy (SLP) techniques developed for WSNs. In this context, the impact of circular and square network models for different network areas, node densities, and radio ranges of sensor nodes on the performance of three popular SLP techniques is investigated. The effectiveness is assessed using performance measures from the body of available literature. It has been found that square deployment performs better for sector-based SLP protocols when network area, node density, and node radio range are held constant. However, circular networks perform better for the same protocol when the radio range is varied beyond a certain threshold while holding all other parameters constant. The trend, however, changes under comparable network settings for the non-sector-based random walk protocols considered in the current work. The results reveal a strong link between deployment geometry and routing logic, demonstrating that sector-based SLP protocols are primarily geometry-sensitive, whereas non-sector-based protocols are more sensitive to radio-range variations. A radio-range threshold of approximately 100 m was identified beyond which privacy gains saturated and energy costs increased. These findings provide practical design guidelines for geometry-aware deployment of privacy-preserving WSNs. However, the conclusions are limited to simulation-based evaluation of three random walk-based SLP protocols in circular and square network topologies, and further validation in irregular and real-world deployments is required.

High-mobility 6G sensing places where strict requirements on waveform robustness are required because the large Doppler shifts can degrade velocity estimation. This paper presents a refined simulation-based comparison of Orthogonal Time Frequency Space (OTFS) and Orthogonal Frequency Division Multiplexing (OFDM) for monostatic velocity sensing at 28 GHz. The evaluation uses a three-way receiver design: the first is OTFS with a single delay–Doppler pilot. The second is a strong full-grid OFDM radar baseline with known time–frequency symbols and range–Doppler matched processing. Third is a communication-oriented sparse-pilot OFDM receiver that reflects practical joint communication and sensing (JCAS) pilot budgets. All branches share identical carrier frequency, subcarrier spacing, grid size, cyclic-prefix length, power normalization, and Monte Carlo settings (250 trials per point). Velocity accuracy is reported through root-mean-square error (RMSE), success rate within ±10 km/h, multipath ablations, a velocity sweep, and quantitative response-concentration metrics. The results show that OTFS with one delay–Doppler pilot achieves near-constant RMSE of about 0.43 m/s and 100% success rate across −10 dB to 25 dB SNR at 120 km/h. A full-grid OFDM radar baseline attains essentially the same accuracy, demonstrating that OFDM is not intrinsically incapable of high-mobility velocity estimation when dense known pilots are available. So sparse-pilot OFDM collapses under the same Doppler conditions (RMSE ≈ 79 m/s and zero success rate) while OTFS preserves compact energy concentration in the delay–Doppler plane. These findings justify OTFS for pilot-efficient high-mobility sensing and clarify that OFDM under-performance in high-mobility sensing can be primarily baseline- and overhead-dependent rather than an absolute waveform verdict.

The IoT (Internet of Things) devices extend the attack surface for cybercriminals, requiring robust IDS (Intrusion Detection Systems). In order to tackle the issues, the AI (Artificial Intelligence), especially the ML (Machine Learning) and DL (Deep Learning) is incorporated into IoT IDS to analyze large datasets, identify complex patterns, and adapt to evolving threats. Hence, the study proposes a modified CNN-LSTM model for real-time intrusion detection and classification in IoT Device Network Logs. The proposed model utilizes the CNNs (Convolutional Neural Networks) for spatial feature extraction and LSTM (Long Short-Term Memory) networks for capturing temporal dependencies, augmenting the accuracy and detection efficiency. The proposed CNN-LSTM model enhances the real-time intrusion detection and contains the ability to detect developing attacks in dynamic IoT environments, which makes it highly flexible for large-scale deployments. The IoT Device Network Logs dataset is used for evaluating the proposed modified CNN-LSTM model. The modified CNN-LSTM model is assessed using the performance metrics such as accuracy, precision, recall, F1-score, time complexity and false alarm rate. As a result, the modified CNN-LSTM model achieves superior performance and earlier detection in contrast to the conventional models deliberating its potential for enhancing IoT security. 

Face recognition systems are increasingly deployed in security-critical applications, but remain vulnerable to presentation attacks such as printed photographs, replay videos, and three-dimensional masks. Although recent face anti-spoofing methods exploit complementary information from RGB, depth, and near-infrared (NIR) representations, existing approaches primarily focus on feature fusion and do not explicitly model the intrinsic consistency relationships among 
these representations. This limitation can reduce robustness against sophisticated spoofing attacks and cross-dataset variations.
This paper proposes a novel Cross-Modal Consistency Learning (CMCL) framework for robust face anti-spoofing using the original RGB input together with RGB-derived depth and NIR representations. The depth and NIR representations are constructed from RGB inputs during the modality preprocessing stage and subsequently processed together with RGB using independent Swin Transformer encoders. A cross-modal attention fusion module adaptively integrates complementary appearance, geometric, and spectral information, while a consistency learning module explicitly encourages feature coherence for genuine samples and emphasizes cross-modal discrepancies associated with spoof attacks. The consistency module is used only during training and removed during inference, avoiding additional deployment overhead.
Extensive experiments are conducted on CASIA-SURF, WMCA, CelebA-Spoof, and MSU-MFSD using intra-dataset, cross-dataset, ablation, feature-space, modality-consistency, and statistical analyses. The proposed CMCL achieves an ACER of 0.70% and accuracy of 99.3% on CASIA-SURF, while achieving an ACER of 2.55% and accuracy of 97.9% on WMCA. The results demonstrate improved spoof detection performance, representation discrimination, and cross-dataset generalization compared with the evaluated state-of-the-art methods. The proposed framework provides a consistency-driven and computationally practical approach for robust face anti-spoofing.

Network port scanning represents a critical reconnaissance phase preceding advanced cyber attacks and poses a significant threat to modern network infrastructures. Conventional signature-based detection systems and static threshold mechanisms are often ineffective in detecting stealthy and low-rate scanning activities in dynamic network environments. This paper proposes a hybrid intrusion detection approach that combines adaptive entropy-based filtering with a Random Forest classifier. The proposed method employs a sliding window mechanism to dynamically adjust detection thresholds based on statistical properties of network traffic, thereby improving adaptability under varying load conditions. Experimental evaluation conducted on the CIC-IDS2017 dataset demonstrates that the proposed model achieves a classification accuracy of 98.7% with a False Positive Rate (FPR) of 1.8%, outperforming both standalone entropy-based and machine learning-based approaches. In addition, the model maintains an average processing latency of 2.3 ms per packet, confirming its suitability for real-time deployment in high-speed network environments. The results indicate that the proposed hybrid approach effectively improves detection performance while maintaining low computational overhead, making it a practical solution for modern intrusion detection systems. However, the proposed approach has certain limitations, including reliance on traffic metadata and reduced effectiveness in encrypted environments.