Protecting Smart-Home IoT Devices From MQTT Attacks: An Empirical Study of ML-Based IDS

Abstract

Smart homes are becoming increasingly popular worldwide, and they are mainly based on Internet of Things (IoT) technologies to enable their functionality. However, because IoT devices have limited computing power and resources, implementing strong security measures is difficult, making the use of intrusion detection systems (IDS) an appropriate option. In this study, we propose an optimized model with high performance for intrusion detection in Message Queue Telemetry Transport protocol (MQTT)-based IoT networks for smart homes. This is done by studying 22 Machine Learning (ML) algorithms based on an extended two-stage evaluation approach that includes several aspects for optimizing and validating the performance to find the ideal model. Based on the empirical evaluation, the Generalized Linear Model (GLM) classifier with the random over-sampling technique produced the best detection performance with 100% accuracy and an f-score of 100%, outperforming previous studies. This study also investigated the influence of automatic feature engineering techniques on the performance of algorithms. With the automatic feature engineering technique, the performance increased by up to 38.9%, and the time required to classify the attacks decreased by up to 67.7%. This shows that automatic feature engineering can improve performance and reduce detection time.