Machine learning-based anomaly detection for smart home networks under adversarial attack

Abstract

As smart home networks become more widespread and complex, they are capable of providing users with a wide range of applications and services. At the same time, the networks are also vulnerable to attack from malicious adversaries who can take advantage of the weaknesses in the network's devices and protocols. Detection of anomalies is an effective way to identify and mitigate these attacks; however, it requires a high degree of accuracy and reliability. This paper proposes an anomaly detection method based on machine learning (ML) that can provide a robust and reliable solution for the detection of anomalies in smart home networks under adversarial attack. The proposed method uses network traffic data of the UNSW-NB15 and IoT-23 datasets to extract relevant features and trains a supervised classifier to differentiate between normal and abnormal behaviors. To assess the performance and reliability of the proposed method, four types of adversarial attack methods: evasion, poisoning, exploration, and exploitation are implemented. The results of extensive experiments demonstrate that the proposed method is highly accurate and reliable in detecting anomalies, as well as being resilient to a variety of types of attacks with average accuracy of 97.5% and recall of 96%.