Smart Home Anomaly Recognition and Prevention (SHARP) : Network Intrusion Detection system utilising Unsupervised Learning and Reinforcement Learning

Abstract

The proliferation of smart home networks consisting of the Internet of Things (IoT) and smart devices has led to a rapidly expanding heterogeneous ecosystem, thereby increasing the attack surface and security vulnerabilities. Whereas traditional supervised Machine Learning (ML) Anomaly-based Network Intrusion Detection Systems (ANIDS) addressed the evolving threat landscape of smart home networks, they face constraints in identifying unknown attacks and lack adaptability in dynamic environments due to the absence of labelled data.

This study reveals a lack of literature concerning ML-based ANIDS utilising unlabelled data while adapting to a dynamic environment. This study introduces a Smart Home Anomaly Recognition and Prevention framework (SHARP), which integrates Unsupervised Learning (UL) and Reinforcement Learning (RL) to identify Distributed Denial of Service (DDoS) cyber-attacks in smart home networks.

SHARP employs an Unsupervised Ensemble (UE) consisting of three unsupervised classifiers to integrate UL and utilises RL’s Exponential weight for Exploration and Exploitation with Experts (EXP4) algorithm to dynamically optimise the UE anomaly score based on varying environmental states. The findings show that SHARP outperforms existing UE methods, achieving an increase of 7.46% in accuracy, 11.91% in precision, 34.27% in recall and a 22.85% increase in the F1-score. SHARP continuously optimises itself by observing the environment without the necessity of labelled data, presenting a promising solution that addresses security vulnerabilities in the rapidly evolving threat landscape of smart home networks.