Enhancing intrusion detection in IoT networks using machine learning-based feature selection and ensemble models

Internet of Things (IoT) technology has evolved significantly, transitioning from personal devices to powering smart cities and global deployments across diverse industries. However, security challenges arise due to diverse devices using various protocols and having limited computational capabilities, leading to vulnerabilities and potential intrusions in IoT networks. This paper addresses the challenge of intrusion detection in IoT by introducing a heterogeneous machine learning-based stack classifier model for IoT data. The model employs feature selection and ensemble modelling to investigate and enhance key classification metrics for intrusion detection of IoT data. This approach comprises two core components: the utilization of the K-Best algorithm for feature selection, extracting the top 15 critical features and the construction of an ensemble model incorporating various traditional machine learning models. The integration of these components harnesses information from selected features and leverages the collective strength of individual models to enhance classification performance. Using the 'Ton IoT dataset,' our experiments compare the ensemble model with individual ones. This research aims to improve key classification metrics for IoT intrusion detection, focusing on accuracy, precision, recall and F1 score. Through rigorous experimentation and comparisons, the proposed ensemble approach showcases exceptional performance, providing a robust solution to fortify IoT network security.