Journal of Security in Computer Networks and Distributed Systems

An Automated Cyber-Physical Smart City Hazard Engine Integrating Deep Learning Computer Vision, Vectorized Non-Maximum Suppression, and Web-GIS Telemetry for Real-Time Road Infrastructure Anomaly Detection and Prioritized Incident Remediation

2026-10-28T09:31:57+00:00

Rapid metropolitan expansion and severe monsoonal weathering inflict acute structural degradation upon subsurface municipal utilities, routinely causing displaced manhole covers, fractured gratings, and exposed stormwater drains. Conventional civic maintenance relies on retrospective complaint queues or manual patrols, creating prolonged intervention delays that precipitate fatal traffic collisions and pedestrian injuries. This study presents PRJ_96 Intel, an automated, cyber-physical hazard surveillance and incident remediation engine developed for smart cities. The platform couples an asynchronous FastAPI backend with an edge-optimized YOLOv8 Nano convolutional neural network, an interactive Leaflet.js Web-GIS spatial telemetry dashboard, and an algorithmic repair prioritization engine. Trained on 4,850 annotated urban streetscape images across metropolitan Bengaluru, India, the lightweight vision detector achieves an overall mean Average Precision (mAP@0.5) of 90.1% while maintaining an inference throughput of 94.3 frames per second. To overcome proposal clustering under irregular pavement textures, a custom NumPy-vectorized Non-Maximum Suppression algorithm is formulated at an Intersection-over-Union threshold of 0.3. Detected hazards are geo-tagged and evaluated through a dynamic Priority Index (0–100) that balances neural detection certainty against localized pedestrian foot-traffic density ratings. Empirical stress testing demonstrates that the platform sustains an end-to-end response latency of 184 milliseconds under concurrent loads, successfully compressing emergency dispatch turnaround from several weeks to under four hours in high-risk zones while preserving privacy through automated visual redaction.

Cooperative Intrusion Response under Noisy Alerts in a Controlled Cloud IoT Network Simulation

2026-10-26T05:49:28+00:00

Automated intrusion response requires an environment in which defensive actions alter subsequent network states. This study evaluates cooperative tabular reinforcement learning in a transparent four-agent cloud IoT simulation with noisy intrusion alerts. Each agent can leave traffic unchanged, inspect it, or isolate its local service. Compromise, propagation, recovery, and action costs are explicitly specified. Independent learners use local alerts, whereas cooperative learners also observe whether a neighbour is alarming and optimise a shared immediate reward. Both methods train for 1,500 episodes on a ring topology and are evaluated for 200 episodes on a star topology across five seeds. Fixed baselines include no intervention, continuous inspection, and isolation on an alert. The alert isolation rule obtains a mean per-agent per-step cost of 0.1569, compared with 0.1793 for cooperative learning and 0.1862 for independent learning. Cooperative learning reduces the compromised fraction to 0.0229, compared with 0.0423 for the rule, but increases isolation to 0.1696 from 0.0855. The measured trade-off therefore does not support a claim of overall superiority for cooperative learning. The results are simulation evidence under declared transition and cost assumptions, not measured containment effects on CICIDS2017 or CICIoT2023 traffic.

Machine Learning-Based Framework for Ransomware Detection and Classification

2026-09-05T07:11:32+00:00

Ransomware has become one of the most dangerous cybersecurity threats, causing significant financial losses and compromising sensitive data across individuals and organizations. Traditional signature-based detection techniques are often ineffective against newly emerging and evolving ransomware variants. Consequently, Machine Learning (ML) has gained considerable attention as an effective approach for identifying ransomware based on behavioral patterns and system activities. This literature survey reviews recent research on ransomware detection using machine learning techniques. It analyzes various detection methods, datasets, feature extraction approaches, and machine learning algorithms such as Decision Trees, Random Forest, Support Vector Machines (SVM), K-Nearest Neighbors (KNN), Naïve Bayes, Artificial Neural Networks (ANN), and deep learning models. The survey compares the performance of these techniques using evaluation metrics including accuracy, precision, recall, and F1-score. Furthermore, this survey highlights the strengths and limitations of existing approaches, discusses the challenges involved in detecting zero-day ransomware attacks, handling imbalanced datasets, and achieving real-time detection. It also identifies current research gaps and explores future directions, including explainable artificial intelligence (XAI), federated learning, and lightweight machine learning models for edge and cloud environments.

A Fault-Tolerant Software-Defined Networking (SDN) Architecture for Reliable Operation of Critical Infrastructure Networks

2026-09-11T05:04:24+00:00

Critical infrastructure networks, including power grids, transportation systems, healthcare facilities, water distribution networks, and industrial control systems, depend on highly reliable and resilient communication infrastructures to maintain uninterrupted operations. As these networks become increasingly complex and data-driven, traditional networking architectures often struggle to provide the flexibility, centralized visibility, and rapid fault recovery required for mission-critical services. Software-Defined Networking (SDN) has emerged as a promising solution by separating the control plane from the data plane, enabling centralized network management, programmability, and dynamic resource allocation. However, the centralized nature of SDN introduces a significant challenge: the controller can become a single point of failure, potentially leading to network-wide disruptions when faults or cyberattacks occur. This paper proposes a fault-tolerant SDN architecture aimed at ensuring the reliable operation of critical infrastructure networks under adverse conditions. The proposed framework incorporates controller redundancy, real-time state synchronization, distributed consensus mechanisms, intelligent failover strategies, and edge-assisted control functionalities to enhance network resilience. These mechanisms collectively enable continuous service delivery during controller failures, link outages, switch malfunctions, and distributed denial-of-service (DDoS) attacks targeting the control plane. The architecture is designed to support rapid fault detection, automated recovery, and seamless traffic rerouting while maintaining network performance and operational stability. The effectiveness of the proposed architecture is evaluated through simulation-based analysis using key performance indicators such as network availability, recovery time, packet delivery ratio, latency, and throughput. Results indicate that the integration of fault-tolerance mechanisms significantly improves network reliability, reduces service disruption, and strengthens the overall resilience of SDN-enabled critical infrastructure environments. The proposed approach provides a practical foundation for deploying dependable, secure, and highly available communication networks capable of supporting the stringent operational requirements of modern critical infrastructure systems.