A Hybrid AI & ML-Based Self-Healing Framework for Internet of Things (IoT) Assisted Smart Wearable to Avoid Security Breaches: Integrating LSTM-Random Forest Detection with DQN-Driven Recovery for Enhanced Network Resilience

Abstract

Internet of Things (IoT) networks are becoming more susceptible to advanced cyber-attacks, including DDoS and ransomware, that are beyond the capability of conventional rule-based intrusion detection systems. This research purpose an AI & ML-based self-healing network model that is constructed on the MAPE-K autonomic structure to automate the entire lifecycle of detection, response and recovery. This architecture uses a hybrid detection model involving Long Short-Term Memory (LSTM) to analyze temporal traffic and Random Forest to classify structured features with a Deep Q-Network (DQN) and Reinforcement Learning (RL) agent to optimize recovery actions. This framework was tested with 3 benchmark datasets CIC IDS2017, UNSW NB15 and Bot-IoT. Experimental findings indicate that the hybrid detection model has an F1-score of 0.91 which is significantly higher than the baselines of single models. Also RL recovery mechanism was used to lower the Mean Time to Recovery (MTTR) to 3.2 minutes which was a 79% reduction of the 15-minute baseline in conventional automated systems. This framework is proven to support real-time inference with latency less than 50 milliseconds and low energy overhead confirmed by simulation experiments using NS-3 and Raspberry Pi 4B edge environments. A holistic measure of resilience provided a total score of 0.92 justifying the capability of the framework to provide network resilience in dynamic and resource-constrained IoT systems.