Cybersecurity risk mitigation and network anomaly detection in smart homes using machine learning and data mining
Abstract
The rapid proliferation of Internet of Things (IoT) devices in smart home environments has significantly increased cybersecurity risks, necessitating the development of intelligent, adaptive, and proactive threat mitigation solutions. Traditional signature-based Intrusion Detection Systems (IDS) and rule-based anomaly detection techniques exhibit significant limitations in detecting zero-day attacks, adversarial intrusions, and evolving cyber threats. To address these challenges, this research proposes a Reinforcement Learning-Based Adaptive Threat Mitigation (RL-ATM) model, designed to enhance cybersecurity risk mitigation and network anomaly detection in smart homes by leveraging reinforcement learning, deep learning, and data mining techniques. Experimental evaluations confirm that RL-ATM significantly outperforms existing cybersecurity solutions, including signature-based IDS, anomaly-based machine learning models, and deep reinforcement learning (DRL) architectures. The proposed model achieved an accuracy of 98.87%, a precision of 97.49%, a recall of 98.36%, and a reduced false positive rate (FPR) of 1.8%, establishing itself as a highly reliable cybersecurity framework for real-world smart home applications. Comparative analysis reveals that traditional IDS models exhibit an accuracy of only 87.42% with an FPR of 6.3%, while anomaly-based ML techniques improve accuracy to 91.15% but still suffer from an FPR of 4.9%. The hybrid CNN + Reinforcement Learning models achieve 92.84% accuracy but lack real-time adaptability to dynamic attack landscapes, making RL-ATM the superior alternative in terms of detection reliability and adaptive response capabilities. This study makes a significant contribution to smart home cybersecurity, providing a highly scalable, adaptive, and autonomous AI-driven security framework capable of mitigating evolving cyber threats in real-time.