Coordinated mission planning for multiple unmanned aerial vehicles in cluttered static three-dimensional environments requires consistent treatment of obstacle-aware motion, fleet-level task allocation, route sequencing, and executable trajectory generation. In many existing approaches, these elements are optimized separately, or fleet-level decisions are made using simplified geometric distances that do not accurately reflect UAV-specific motion feasibility in obstacle-constrained space. This paper presents a Multi-UAV Planning Framework for Task Allocation, Route Optimization and Trajectory Smoothing for static environments with known obstacle geometry. In the first stage, an offline single-UAV planner based on a hybrid Differential Evolution and Enhanced Whale Optimization Algorithm computes feasible raw paths for all relevant ordered node pairs and constructs a UAV-specific directed travel-cost matrix. In the second stage, these planner-derived matrices are used for feasibility-aware balanced task distribution and route optimization with exchange-based refinement under a composite total-cost–makespan objective. In the third stage, the raw paths corresponding to the final selected routes are reconstructed and transformed into executable trajectories by adaptive cubic B-spline smoothing. Experimental evaluation was conducted at the local-planning, fleet-planning, and smoothing levels in known static environments. The hybrid planner generated high-quality pairwise obstacle-avoiding paths and exhibited favorable convergence behavior relative to standard WOA, PSO, DE, SOS, and GWO in the tested scenarios. At the fleet level, the full framework reduced makespan by 2.1–3.4% and the composite objective by 0.9–1.4% relative to balanced partitioning without exchange refinement on benchmark instances. In the smoothing stage, the adaptive cubic B-spline reduced path length by 8.5% and maximum curvature by 41.5% relative to the unsmoothed polyline representation. These results demonstrate that the proposed hierarchical formulation is computationally effective, physically consistent, and well suited to multi-UAV mission planning.

This paper presents a machine learning-based model designed to forecast trends in enrollment in the Department of Computer Studies at NEMSU–Cantilan Campus, specifically for the Bachelor of Science in Computer Science (BSCS), Bachelor of Science in Computer Engineering (BSCpE), and Bachelor of Science in Information Technology (BSIT) programs. Twenty semesters (2015–2025) of historical program-level enrollment were chronologically split, with the most recent semesters withheld for validation, and a lagged feature set (Lag 1–Lag 3) was constructed for an Artificial Neural Network (ANN) and benchmarked against Holt-Winters Exponential Smoothing and ARIMA, both tuned on the same training split. Employing the ANN model, this study achieved an aggregate Mean Absolute Percentage Error (MAPE) of 3.62%, outperforming the Holt-Winters (25.59%) and ARIMA (31.09%) baselines on this dataset, a ranking corroborated by RMSE and MAE (Table 6) and confirmed against a small LSTM and Prophet, neither of which outperformed the ANN at this sample size; a per-program breakdown (Table 7), however, shows this advantage is not uniform, ranging from 3.93% MAPE for BSCS to 13.48% for BSIT. The enrollment projections indicate consistent growth across all programs. The BSCS program is likely to increase from 297 students in 2025 to 401 by 2028, exhibiting a 35% growth. The BSIT program is projected to experience the most significant expansion, growing from 1,078 students in 2025 to 2,714 students by 2028-an increase of 151%, a trend consistent with the program’s sharp historical acceleration after 2022 and the recursive nature of the multi-step forecast, which compounds this recent growth forward. Meanwhile, the BSCpE program is expected to grow more gradually, from 153 students in 2025 to 186 by 2028, showing a 21.6% increase. Overall, the total enrollment in the Department of Computer Studies is expected to rise from 1,517 students in the second semester of 2025 to 3,031 by the first semester of 2028, marking a 100% increase. These findings highlight the accuracy and adaptability of ANN-based models in capturing nonlinear trends in enrollment, offering a valuable tool for strategic planning, faculty and classroom allocation in state universities and colleges in the Philippines.

Malicious user detection in spectrum sensing is a critical challenge in Cognitive Radio Networks (CRNs). Traditional rule-based mechanisms lack adaptability to dynamic behaviors, while existing AI techniques often overlook scalability and real-time constraints. This paper proposes a novel hybrid framework that integrates adaptive trust-based mechanisms with AI-powered anomaly detection and blockchain technology to achieve superior detection accuracy (>90%), energy efficiency (30% reduction), and scalability (supporting 500+ nodes with blockchain throughput >750 transactions/second). The framework dynamically updates trust scores using machine learning models and leverages blockchain for secure and transparent spectrum management. Comparative simulations demonstrate superior performance compared to existing methods. The proposed methodology addresses the limitations of static trust mechanisms and offers a robust solution for real-time malicious user detection in CRNs.

The recent development of the Internet of Things (IoT) has increased the severity of security threats. It is mainly caused by IoT devices' inherent weaknesses, making them vulnerable to attack. Therefore, strengthening the security of such network systems is crucial. This study proposes a novel stacking model to identify attacks in the IoT environment. As a first step, we preprocess the data to make it more reliable. To address the issue of class imbalance, synthetic minority samples are generated by using SMOTE-SVM. Then, a novel stacking model was built by integrating four neural networks: deep neural network (DNN), recurrent neural network (RNN), long short-term memory (LSTM) and gated recurrent unit (GRU) with hyper-parameter tuned Light gradient boosting machine (BTE-LGBM). The performance of the proposed stacking model was evaluated on two recent IoT datasets, namely the ToN_IoT and CIC-IoT23. The efficacy of the suggested stacking model is evaluated and compared with Deep learning, machine learning, and state-of-the-art approaches with respect to metrics such as detection rate, precision, accuracy, and F1 score. The findings of our experiments indicate that the suggested IDS achieves a high accuracy of 99.81% and 99.78% for ToN_IoT and CIC-IoT23 datasets, respectively. It might enhance IoT device security, eventually benefiting consumers who depend on these devices. These findings, however, are based on benchmark dataset and could be impacted by variables including class distribution, attack diversity, and dataset characteristics. More research is needed to determine how well the model performs in real-world IoT contexts with changing attack patterns and heterogeneous devices.

Wireless Sensor Networks (WSNs) are the core of Internet of Things (IoT) applications as they enable energy-efficient and scalable real-time data acquisition. However, WSN-based IoT systems have been facing great challenges in achieving energy efficiency, security and scalability especially in applications like smart agriculture. To tackle these problems, this study proposes a new framework called OptiBC-WSN that combines Particle Swarm Optimisation (PSO) clustering, AES-128 encryption, Proof of Authority (PoA) blockchain and Random Forest-based Intrusion Detection System (IDS). We have implemented PSO for energy efficient clustering, PoA for trust management and a Random Forest based IDS for attack detection on Intel Berkeley and GreenOrbs datasets. OptiBC-WSN consumes 0.25 mJ/node energy, LEACH consumes 5 mJ/node and K-Means+Bee consumes 0.3 mJ/node. It also achieves a breach rate of 0.13% vs. 2% for LEACH and 0.2% for K-Means+Bee and scales to 10,000 nodes with 15,950 bytes overhead (vs. 20,000 bytes for LEACH). Its IDS has 90% accuracy to DDoS, Sybil and Jamming attacks with fairness index 0.92. We perform scale-up tests from 100 to 10K nodes with GreenOrbs dataset and synthetic topology generation, and validate core energy and security metrics with the Intel Berkeley dataset (54 nodes). In smart agriculture it cuts irrigation by 15% and CO2 emissions by 60 kg/1,000 nodes/year. Future work will explore the use of blockchain sharding and adaptive IDS to improve the scalability and security of IoT.

Wireless Mesh Networks (WMNs) provide low-cost, self-organizing and self-healing connectivity, but multi-hop interference, hidden-node effects and unbalanced load make delay-sensitive Voice over Internet Protocol (VoIP) communication difficult to support. This paper presents a Particle Swarm Optimization (PSO)-driven node monitoring and traffic scheduling framework for QoS-aware VoIP in WMNs. In the revised method, multi-level monitoring is explicitly defined through three measurable levels: node-state monitoring (identifier, residual energy and queue occupancy), link-quality monitoring (delivery probability, loss, delay and interference), and traffic/QoS monitoring (VoIP classification and priority scheduling). These normalized features are combined in a dimensionally consistent PSO fitness function that jointly maximizes packet delivery ratio and residual energy while minimizing delay, packet loss and hop count. A MATLAB-based discrete-event simulation was conducted for WMNs with 50-300 nodes, bidirectional CBR/UDP VoIP flows, IEEE 802.11 CSMA/CA access and common channel/interference assumptions. Under the modeled conditions, the proposed PSO-MLNM-EPDR method achieved 97.7-98.8% packet delivery ratio, compared with 94.1-95.2% for RAAOR-WMN and 92.7-93.7% for FDOE-WMN, while keeping one-way delay within 7.4-8.6 ms. The study is limited to controlled simulation settings without mobility or field deployment; therefore, future work should validate the method under realistic traffic bursts, mobility and heterogeneous radio environments.

Intrusion Detection Systems (IDS) are essential for protecting modern networks against unauthorized access and evolving cyber threats. A persistent challenge in IDS design is the high dimensionality of network traffic data, which complicates the identification of the most relevant features for effective detection. This study introduces a modified multi-stage ensemble feature selection (MMSE-FS) framework that incorporates algorithmic adaptations of Random Forest (RF), Principal Component Analysis (PCA), and KBest methods. These enhanced variants are integrated through an intelligent ensemble voting mechanism, followed by a refinement stage that further strengthens feature relevance and discriminative capability. 
To validate the proposed framework, experiments were conducted on the UNSW-NB15 benchmark dataset, reducing 49 initial features to 18 critical ones. The dataset was partitioned into 70% training and 30% testing subsets, and classification performance was evaluated using five machine learning classifiers (DT, RF, GB, KNN, and LR). Key hyperparameters of the proposed MMSE-FS framework (α = 0.75, λ = 1.0, and B = 50 bootstrap repetitions) were determined through 5-fold cross-validation on the training partition and subsequently fixed for all experiments. The proposed framework achieved detection accuracies ranging from 99.03% to 99.83% for binary classification and from 94.20% to 96.60% for multi-class classification.
Compared with conventional feature selection methods, the proposed MMSE-FS framework substantially reduced the feature space while maintaining high detection performance across both binary and multi-class intrusion detection tasks. The reported results were obtained using the UNSW-NB15 dataset following the adopted preprocessing strategy, which excluded extremely underrepresented attack classes.

Distributed Denial of Service (DDoS) attacks have emerged as one of the most critical cybersecurity threats to cloud computing environments, significantly affecting service availability, resource utilization, and operational reliability. The dynamic nature of cloud traffic and the increasing sophistication of attack patterns make accurate and efficient DDoS detection a challenging task. To address this issue, this paper proposes a hybrid feature-selection and classification framework, namely Random Forest-based Oppositional Crow Search Algorithm (RF-OCSA), for intelligent DDoS attack detection in cloud environments. The proposed framework consists of data preprocessing, optimal feature selection using the Oppositional Crow Search Algorithm (OCSA), and attack classification using the Random Forest (RF) classifier. OCSA employs oppositional learning to enhance search diversity and identify the most discriminative features while reducing feature redundancy and computational overhead. The effectiveness of the proposed RF-OCSA model is evaluated using four benchmark datasets, namely CIC-DDoS2019, AISED, TestCloudIDS, and UNSW-NB15. Experimental analysis is conducted using Accuracy, Sensitivity, Specificity, Precision, and F-measure as evaluation metrics, while NS-3 is utilized to emulate representative cloud-network traffic scenarios. The results demonstrate that RF-OCSA consistently outperforms GHLBO, CNN-LSTM, and DeepDefend by achieving superior detection performance across all datasets, with classification metrics exceeding 98% in most cases. The integration of oppositional feature optimization and ensemble learning significantly improves attack detection capability, reduces false alarms, and enhances model robustness. These findings indicate that RF-OCSA is an effective and scalable framework for securing cloud infrastructures against evolving DDoS attacks.

Opportunistic Internet of Things (O-IoT) networks operate in highly dynamic, infrastructureless environments where connectivity is intermittent and unpredictable, making efficient and reliable data delivery a persistent challenge. Traditional routing protocols such as Epidemic Routing and PRoPHET have been widely studied, yet both present significant drawbacks: Epidemic Routing ensures high delivery probability by replicating messages extensively, but this causes excessive buffer usage, bandwidth consumption, and energy drain, while PRoPHET employs probabilistic forwarding based on encounter histories, which is more resource-efficient but struggles in highly mobile or sparse networks where prediction accuracy decreases. To overcome these issues, this paper proposes a Hybrid Adaptive Routing framework that integrates the predictive capability of PRoPHET with a controlled epidemic fallback mechanism. The framework applies a predictability threshold of 0.6 to decide when to rely on probabilistic forwarding and when to activate epidemic replication, while carefully constraining the latter with EPIDEMIC_LIMIT = 5, HOP_LIMIT = 8, and TTL = 300 minutes to prevent resource exhaustion. . The framework was subsequently simulated and analyzed in an Opportunistic Network Environment (ONE) at different network densities and compared with the traditional routing protocols Epidemic and PRoPHET. The system's performance was evaluated using parameters such as delivery probability, overhead ratio, average latency, hop count, and buffer utilization. Experimental results confirm the approach’s effectiveness at high node density (246) nodes, where the hybrid protocol achieves a 19.03% improvement in delivery probability over Epidemic routing and 30.45% improvement over PRoPHET, alongside a 43.5% and 31.1% overhead reduction compared to Epidemic and PRoPHET respectively, and 35.5% and 37.1% latency reduction compared to Epidemic and PRoPHET respectively, making it a robust and resource-efficient solution for real-world O-IoT applications.

With the rapid adoption of smart home solutions and related technologies, edge computing has emerged as a key enabler by offering low-latency data processing, increased efficiency and improved scalability. However, this integration in IoT systems introduces complex security challenges in smart home edge environments, increasingly susceptible to cyber threats such as denial-of-service (DoS), malware injection, passive surveillance, and unauthorized access. This paper investigates intelligent intrusion detection and attack classification strategies specifically designed for smart home edge systems. Using the comprehensive ML-EdgeIIoT dataset, this study designs and evaluates a machine learning-based intrusion detection framework for multiclass classification of eight categories of IoT network attacks, namely Backdoor, MITM, DDoS, Ransomware, Password Attack, SQL Injection, Prob-attacks, and Normal traffic while minimizing false positives and false negatives. The framework incorporates data cleaning, correlation- and feature importance-based feature selection, hyperparameter optimization using gridsearchCV, model training, and ensemble learning. A set of machine learning models comprising Artificial Neural Network, Balanced Random Forest, K-Nearest Neighbours, Random Forest, and Logistic Regression was implemented and comparatively evaluated. Two ensemble techniques were subsequently developed using the three best-performing classifiers: (1) a stacking ensemble with Logistic Regression as the meta-learner and (2) a Top-3 majority voting ensemble. Model performance was evaluated using accuracy, precision, recall, F1-score, confusion matrix, and ROC-AUC. Robustness and generalization of the individual machine learning models were assessed through stratified 10-fold cross-validation for the three best-performing classifiers. The Top-3 voting ensemble subsequently achieved the highest performance on the independent test set, with accuracy of 99.24%, average precision of 98.75%, recall of 99.00%, and an F1-score of 99.00% for all attack classes, while reducing misclassification compared with individual classifiers. The findings of this study significantly enhance the understanding of smart home edge computing security, which will pave the way for more robust and intelligent threat detection frameworks.

This paper addresses the problem of efficient data transmission under dynamically changing network and computational conditions by proposing an adaptive data compression method based on context-aware selection of compression algorithms and their parameters. Unlike conventional static approaches, the proposed method performs real-time analysis of data characteristics, network bandwidth, latency, and available computational resources, enabling dynamic selection of the optimal compression strategy through multi-criteria optimization. The scientific novelty of the work lies in the integration of data-driven and environment-aware adaptation within a unified decision-making framework that simultaneously minimizes end-to-end transmission delay while balancing compression ratio and processing overhead. Experimental evaluation was conducted using both synthetic datasets and the standard Silesia Corpus benchmark. The synthetic datasets included repetitive low-entropy text (repeated.txt), structured JSON data (structured.json), moderate-complexity text (example.txt), and high-entropy binary streams (random.bin), representing realistic web content and raw data transmission scenarios. The Silesia Corpus, containing approximately 200 MB of heterogeneous real-world files, including text, binaries, and images, was used for validation and benchmarking. The proposed method was evaluated using compression algorithms such as LZ4, Zstandard, Brotli, and ZSTD under different network conditions and system loads. Experimental results show that the adaptive approach reduces total transmission time by an average of 23%, improves compression efficiency by 16%, and decreases computational resource consumption by 13% compared to conventional static compression methods. The software implementation is based on a modular service-oriented architecture that supports real-time monitoring, dynamic algorithm switching, and scalable deployment in distributed, cloud, streaming, and Internet of Things environments.

Insider attacks pose a significant security threat precisely because they originate from individuals with authorized access, making them inherently difficult to detect and prevent; addressing this issue is crucial for preserving the confidentiality, integrity, and availability of organizational systems. This paper contributes to mitigating insider attacks by proposing an approach called RAFIA, which monitors the system, evaluates the risk of insider threats, and blocks malicious actions before unauthorized or high-risk access is granted. The security policy is specified using an enhanced version of Linear Temporal Logic, called Risk-LTL, which evaluates the risk of each new action based on system history, including log files or traces, and a risk evaluation function provided as input. Risk evaluation is based on combining maliciousness probability and impact assessment, enabling quantitative estimation of the risk associated with user actions and action traces. Access decisions are governed by configurable risk thresholds specified within Risk-LTL policies. To strengthen decision-making, the model frames access control as a game between users and the organization. By applying game-theoretic tools, the system analyzes user behavior and makes access decisions that discourage malicious actions and reward honest ones. The objective is to reach a Nash equilibrium, where both players act rationally and securely. The proposed approach aims to improve the effectiveness of access control by reducing dishonest behavior and promoting more stable, risk-aware system interactions. Experimental evaluation using synthetic workloads of up to 10,000 access requests demonstrated the practicality of the proposed framework. RAFIA achieved an average authorization latency of approximately 3.3 ms while improving the F1-score compared with a conventional static-threshold access-control baseline.

Financial fraud presents a major challenge to financial establishments, with Nigerian banks losing over ₦685 million to digital fraud in 2023. Traditional rule-based detection systems have high false-positive rates and limited adaptability; meanwhile, existing machine learning models are generally trained on non-localized datasets that ineffectively represent African fintech ecosystems. This study proposes a Hybrid Stacked Ensemble framework for fraud detection that improves detection accuracy, robustness, and explainability in localized financial environments. The proposed framework combines Random Forest, Gradient Boosting, and Extra Trees as base learners with XGBoost as the meta-classifier and integrates SHAP for model explainability. Performance was evaluated using the Kaggle credit card fraud dataset (284,807 transactions; 0.17% fraud) and a newly curated Nigerian synthetic dataset (150,000 transactions; 1.12% fraud) incorporating localized fraud patterns such as POS, USSD, mobile money, and rural–urban transaction disparities. Class imbalance was addressed using SMOTE oversampling, random undersampling, and cost-sensitive learning. On the Kaggle dataset, the Hybrid Ensemble achieved 99.96% accuracy, 97.14% precision, 79.70% recall, an F1-score of 0.880, and an AUC-ROC of 0.986, outperforming the best individual classifier in recall and AUC-ROC. On the Nigerian dataset, where individual classifiers achieved recall below 3.1%, the proposed framework attained 64.10% recall, an F1-score of 0.460, and an AUC-ROC of 0.866, representing improvements of 106.77% in recall and 488.57% in F1-score over XGBoost. Ablation studies and paired t-tests (p < 0.001) confirmed the effectiveness of the stacking strategy. The study contributes a localized Nigerian fraud dataset, a hybrid stacked ensemble architecture that exploits classifier diversity for improved fraud detection, and an explainable AI framework that enhances transparency, accountability, and regulatory compliance. Deployment as a containerized Streamlit application with JWT authentication demonstrates the framework's practicality as a scalable, explainable, and deployment-ready solution for fraud detection in Nigeria and similar African financial ecosystems.

The longevity of the network depends on making the correct route decisions. The term "optimal routing" in wireless sensor networks refers to selecting the best path among the available routes in order to reduce energy consumption and increase network lifetime. Numerous optimization algorithms considering parameters like energy remaining in a node, node distance, network topology, and link quality when choosing the best routes. In order to enable data aggregation and energy-efficient routing through the CH, optimal routing techniques are typically implemented in cluster-based WSNs. In Heterogeneous WSNs (HetWSNs), reliability studies pertain to the network's capacity to deliver information to the base station. Extreme weather conditions, sensor node battery depletion, hardware and software malfunctions, and other factors all have an impact on a heterogeneous WSN's reliability. This article examines the application of optimal routing to both chain-based HetWSNs and also focuses on reliability studies in HetWSNs with chain-based connections that use optimal routing. For five levels of energy-based HetWSNs, an algorithm known as ORB-PEGASIS (Optimal Routing Based Power Efficient Gathering in Sensor Information System) is created. Our suggested ORB-PEGASIS algorithm increases the network lifetime of the HetWSNs by, according to a comparison of the obtained results with various algorithms, including PEGASIS-E by 68.42%, PEGASIS by 112.8%, PEG-ACO by 60.43%, and IEEPB-KMO by 58.47%.

Dynamic spectrum access (DSA) in 5G IoT setups with cognitive radio is characterized by rapid and decentralized decision-making processes in highly non-stationary wireless environments, limited communication needs, and restrictive bounds. In this work, we present F-DMRL, a federated, communication-efficient decentralized meta-reinforcement learning framework for allowing a massive number of IoT devices to meta-learn collectively about spectrum-access strategies in a decentralized way without centralized control and without an extensive amount of inter-agent communication. Our method incorporates lightweight federated meta-parameter aggregation with gradient sparsification and periodic communication, allowing devices to only compress the meta-updates during this process and then adapt locally for task-specificity. We have presented analytical speedup guarantees and upper bounds on communication cost under bounded environmental drift and shown that using the approach proposed here, F-DMRL preserves convergence properties while posing a large reduction in coordination overhead at the same time. Simulations across various 5G IoT spectrum environments showed that F-DMRL performed faster adaptation (up to 45% fewer episodes), higher spectral efficiency, and lower interference probability compared to centralized meta-RL, federated DRL, and traditional decentralized RL baselines. Simulation results averaged across 10 independent runs demonstrate improvements of 45% faster adaptation and 60–80% lower communication overhead relative to baseline methods, while maintaining stable convergence.

Continuous Integration and Continuous Deployment (CI/CD) pipelines have become fundamental to modern software engineering, enabling rapid and reliable delivery of applications. However, their automation introduces critical vulnerabilities, particularly credential leaks and misconfigurations, which undermine the security of development and deployment environments. This study investigates security risks in Dock-er-based GitHub Actions workflows and proposes a tailored, DevSecOps-aligned security checklist to mitigate these threats. A systematic literature review was combined with hands-on experiments, in which controlled credential exposures and workflow misconfigurations were deliberately introduced and analyzed. Security controls such as secret scanning with GitGuardian and TruffleHog, configuration validation with GHAST, and access control enforcement were tested in a CI/CD testbed. The findings demonstrate that these integrated methods significantly reduce the risk of credential leakage and pipeline hijacking, while maintaining minimal performance overhead. The novelty of this work lies in consolidating fragmented best practices into a work-flow-specific model that is immediately applicable to real-world projects. This contrib-utes actionable guidance for secure-by-design CI/CD pipelines, offering practical protection against supply-chain threats while preserving delivery speed and scalability.

Distributed Denial of Service (DDoS) attacks have gained popularity among cybercriminals as a favoured method of disruption. Application layer DDoS attacks are particularly intricate, as they overload web servers with re-quests, rendering them inaccessible to legitimate users and causing availability issues. These attacks are challenging to detect through network and transport-level security measures, making them even more concerning. This paper explores various categories of DDoS attacks, encompassing volumetric and protocol-focused attacks, with a particular focus on application-layer attacks, classifying them into Protocol-specific attacks and Generic attacks. It also delves into diverse defence strategies tailored to combat related attacks, such as HTTP Flood, DHCP starvation, SlowLoris, and others. Unlike earlier surveys, which centre on vulnerability-oriented taxonomies through 2017–2020, this work introduces an explicit, criterion-based comparative evaluation framework for attacks and defences, and extends the taxonomy with post-2020 developments containerized and cloud native low-rate attack surfaces, machine learning-driven detection and adversarial evasion, and zero-trust-based mitigation illustrated with the 2023 HTTP/2 ‘Rapid Reset’ incident. Finding that detection-only mechanisms still dominate current defences, the paper identifies recurring bottlenecks and proposes con-crete future-research directions, including detection resistant to adversarial machine learning and low-rate attack detection in containerized and serverless environments.

Traditional signature-based Web Application Firewalls (WAFs) have difficulty detecting increasingly complex assaults that target web applications, such as SQL injections, Cross-Site Scripting (XSS), and API misuse. In this study, we introduce LLM-WAF, a new intelligent firewall architecture that uses Large Language Models (LLMs) to analyze HTTP traffic contextually and semantically. Our framework integrates pre-trained language models with realtime traffic monitoring pipelines to identify malicious payloads through natural language processing capabilities rather than static rule matching. The system incorporates a continuous learning mechanism using reinforcement signals from detected attacks to adapt to emerging threat vectors automatically. In comparison to conventional WAF systems, experimental evaluation on benchmark datasets such as the CSIC 2010 HTTP Dataset and real-world traffic scenarios shows that LLM-WAF achieves 96.8% detection accuracy with an F1=0.95cand dramatically lowers false positives.

With the accelerated proliferation of cloud services and web-based applications, exposure to sophisticated cy-ber threats like zero-day vulnerabilities, advanced persistent threats, and application-layer attacks, has sharply increased. Conventional intrusion detection systems, along with cryptographic security mechanisms, often do not fulfill the require-ments of adaptive detection, privacy preservation, and variety of scalability within distributed systems. To alleviate these problems, this paper suggests a cross-layer adaptive security model, which includes, in the secure cloud and web applica-tions, machine learning-based anomaly detection complemented with advanced cryptographic security. The model com-bines, in this context, lightweight local anomaly detection, federated learning, selective privacy, and a deep reinforcement learning-based threat detection and security orchestration. Through federated learning, active participation in the learning process is assured, while the selective privacy mechanism preserves the model parameters. The deep reinforcement learn-ing agent adjusts the interaction, aggregation, and privacy settings according to demands of the adaptive system and the environment. The assessment of the model is realized through the CSE-CIC-IDS2018, CIC-IDS2017, and the CSIC 2010 HTTP benchmark datasets to verify the model in detection, generalization, and operational effectiveness. The results of the performed tests reflect an accuracy of 97.9%, a F1 score of 97.5%, a false positive rate of 1.8%, and a detection latency of 36 ms, surpassing performance of conventional federated and centralized state-of-the-art models. Cross-dataset testing validates the model effectiveness in the presence of highly variable traffic. The results suggest that adaptive orchestration, federated learning, and selective privacy preservation, when combined, substantially boost intrusion detection, decrease communication overhead, and ensure privacy preservation. Therefore, this framework is a scalable and robust approach to intrusion detection within contemporary cloud and web settings.

In this paper, we address the problem of estimating the distance between a camera and a photographed object using minimal prior information. Specifically, the goal is to obtain distance estimates without access to intrinsic camera parameters such as focal length, sensor size, or lens distortion coefficients. We propose three simple heuristic methods, which can be viewed as lightweight variants of the classical camera pinhole model (CPH), but do not require calibration. Instead, they rely only on the approximate real height of a known object in the scene. The methods were validated on a representative dataset of images captured with several modern cameras and smartphones, using known object dimensions as ground truth. Their performance was compared against CPH using error metrics such as mean absolute error (MAE), mean absolute percentage error (MAPE), root mean squared error (RMSE), mean signed error (MSD), coefficient of determination R2), and supported by statistical testing (Shapiro–Wilk, ANOVA, Kruskal–Wallis). The analysis confirmed that, while less precise than fully calibrated approaches, the proposed heuristics achieve consistent and reliable distance estimates under minimal assumptions. These methods are particularly suited for lightweight applications and devices with fixed focal lengths, such as smartphones.

Black-hole (BH) attacks are among the most critical security threats in Mobile Ad Hoc Networks (MANETs) due to their decentralized and highly dynamic nature. Existing protection mechanisms primarily rely on threshold-based monitoring, trust evaluation, and route observation techniques, which often suffer from limited packet information, poor scalability, and inadequate attack verification capabilities. To address these limitations, this paper proposes a centralized federated framework for black-hole attack detection in MANETs. The proposed framework integrates Linear Scaling-based Shark Smell Optimization Algorithm (LS-SSOA) for Cluster Head (CH) selection, Gini Canberra Spearman-based K-Means Algorithm (GCS-KMA) for clustering, Round Log Sum BCrypt-based Message Authentication Code (RLBC-MAC) for secure route authorization, and Judy-Verkle Tree (JVT) for malicious path verification. A federated learning architecture consisting of a global server and local models is employed to continuously monitor network activities and improve attack detection efficiency. Experimental evaluation was conducted using MANET networks comprising 50–250 nodes in a Python-based simulation environment. The proposed framework achieved a detection accuracy of 98.96%, a false prediction rate of 2.06%, throughput of 6445 kbps, and reduced computational overhead compared with existing methods. The results demonstrate that the proposed centralized federated framework significantly improves network security, route reliability, and attack detection performance in MANET environments.

Abstract: Mobile Ad Hoc Networks (MANETs) are autonomous wireless networks that do not rely on fixed infrastructure for communication among the mobile nodes. The mobile nodes communicate with each other with no centralized control, and routes are not established in advance. This makes MANETs vulnerable to routing attacks such as Black Hole and Sybil attacks, which divert or drop packets and thereby degrade network performance. This paper presents a novel hybrid approach combining Federated Learning (FL) and Graph Attention Network (GAT), which is termed as FLAT. GAT dynamically assigns importance to neighbouring nodes, allowing the model to capture the mobility, link instability, and heterogeneous node behaviour inherent in MANETs. FL, in turn, enables routing and attack patterns to be learned in a decentralized manner without sharing raw data. The approach was evaluated in NS-3.36 on networks ranging from 10 to 150 nodes, under Black Hole and Sybil attacks with 20% of the nodes acting maliciously, and was compared against AODV, SAODV, AOMDV, and the optimization-based Dolphin Cat Optimizer, using PDR, PLR, throughput, and end-to-end delay as evaluation metrics. It is found that the PDR of FLAT is approximately 15.1% higher than AODV and 8.2% higher than the Dolphin Cat Optimizer, while its PLR is reduced by about 77% relative to AODV, 75% relative to SAODV, and 50% relative to the Dolphin Cat Optimizer. With respect to throughput performance, FLAT shows a performance level of about 4.4, 2.8, 2.3, and 2.1 times higher than AODV, SAODV, AOMDV, and Dolphin Cat Optimizer respectively. The above findings clearly show that FLAT outperforms current techniques with regard to all measured parameters.

Wireless Sensor Networks (WSNs) play a critical role in various applications, including environmental monitoring, healthcare, and industrial automation. However, these networks face significant challenges related to energy efficiency, fault tolerance, and reliable data transmission, particularly in dynamic environments. Existing clustering and routing techniques often fail to ensure seamless fault tolerance and energy optimization simultaneously. Many traditional approaches lack robust mechanisms to handle Cluster Head (CH) failures, resulting in reduced network stability and shorter operational lifetimes. To address these limitations, this study proposes a Fault-Tolerant Backup Cluster Head with Ant Colony Optimization (FT-BKCH-ACO) approach that enhances energy efficiency and network resilience. The methodology involves optimized CH and Backup CH (BKCH) selection, considering parameters such as residual energy, distance to the base station, and network density. Additionally, Ant Colony Optimization (ACO) is employed to dynamically adjust pheromone levels for energy-efficient routing, ensuring reliable intra-cluster and inter-cluster communication. Simulation results demonstrate that the FT-BKCH-ACO approach significantly improves energy consumption by 23.2%, packet delivery ratio by 10.5% and end-to-end delay by 17.8% compared to existing models. The inclusion of backup CHs ensures seamless communication even in the event of node failures, making this method highly suitable for IoT-enabled WSN applications. The proposed approach bridges the gap between fault-tolerant clustering and adaptive routing, offering a scalable and energy-efficient solution for large-scale sensor networks.

Antenna selection in reconfigurable intelligent surface (RIS)-assisted multiple-input multiple-output (MIMO) systems presents has significant computational challenges.  The contribution of each transmit antenna is determined by the combined effects of direct and RIS-reflected channels. To address the complexity of combinatorial search a graph neural network (GNN)-based antenna selection framework is proposed. In this framework transmit antennas are represented as graph nodes with channel-correlation information forming. The graph edges and magnitude-phase channel statistics serve as node features. A three-layer feedforward GNN is trained using greedy-selection labels generated from 1000 channel realizations and evaluated on 200 independent test realizations. For a 16×8 MIMO system assisted by a 64-element RIS at 28 GHz the proposed method achieves a spectral efficiency of 63.41 bits/s/Hz and corresponding to 94.5% of the greedy baseline performance of 67.09 bits/s/Hz. While reducing the average selection time from 0.441 ms to 0.025 ms. These results determine that graph-structured learning enables near-greedy antenna selection with considerably lower inference complexity. The current study is limited to simulated settings with fixed system dimensions and idealized channel assumptions; future work will address broader channel models and larger-scale configurations.

Cloud schedulers that pair workload prediction with reinforcement learning (RL) rarely check whether a given prediction can actually be trusted, and earlier confidence-gated designs often mix current and future information inconsistently. We fix that inconsistency and build a causally consistent, confidence-gated LSTM-DQN scheduler: an LSTM forecasts next-step workload, a retrospective, error-based confidence score gates how much a Deep Q-Network (DQN) scheduler leans on that forecast, and only information available at decision time is ever used. We implement and pilot-test this architecture in a Python-based discrete-event simulation configured to match a CloudSim-style environment (10 hosts, 30 VMs), benchmarking it against FCFS, Round Robin, standard RL, two ablation variants, and two simplified state-of-the-art comparators across five random seeds. The results show the method works as intended: it trains stably and safely on every seed, holds response time and SLA violations in line with standard RL and simple heuristics, and clearly outperforms a metaheuristic-augmented Q-learning baseline, which suffered severe instability under the same conditions. Code, raw results, and statistical tests are released for independent verification, with scaled-up training identified as the natural next step to test whether larger performance gains emerge.