AI-Driven Self Healing Network for Link Failure, Detection and Recovery

Wired communication networks form the backbone of modern information systems, enabling high- speed, low-latency data exchange for mission-critical applications in industries such as finance, healthcare, manufacturing, and defense. Despite their reliability and bandwidth advantages, wired networks remain susceptible to link-level failures arising from hardware degradation, congestion, and misconfiguration. These failures often lead to packet loss, increased latency, service downtime, and in some cases, a complete network partition. Traditional recovery methods, such as static failover or manual intervention, are insufficient in complex and dynamic network environments where quick reaction times are critical. This paper presents a comprehensive AI-driven self-healing network model specifically tailored for wired network infrastructures. The approach is implemented and evaluated using OMNeT++ 6.1 and the INET 4.5 framework. It combines real-time fault detection, machine learning-based decision support, and automated rerouting mechanisms to minimize the impact of link failures. A hybrid mesh- star topology was chosen to emulate enterprise-level networks with redundant backbone connections and distributed access layers. The simulation incorporates realistic UDP traffic flows, mid-simulation link failures, and intelligent recovery logic triggered by anomaly detection. Machine learning models, including Random Forest and XGBoost, are trained on network telemetry to identify and react to abnormal patterns indicating potential failures. Experimental results demonstrate the effectiveness of the proposed solution in significantly reducing packet loss, improving recovery time, and maintaining throughput stability. The system dynamically adapts to changing conditions without the need for pre-configured backup routes. This research highlights the feasibility and practical benefits of integrating AI into wired network fault management, paving the way for future implementations in real-world enterprise and data center environments.