An intrusion detection system for industrial IoT using chi-square feature selection

Abstract

The expansion of the Industrial Internet of Things (IIoT) has led to advancements in various industries, but that has also exposed infrastructure which is critical to increasing cyber threats. In response, this paper addresses the important role of feature selection in improving the performance of Intrusion Detection Systems (IDS). Utilizing the Edge-IIoTset, this paper introduces a model which includes data pre-processing, feature selection using chi-square method, intrusion detection using machine learning and performance analysis. The experimentation reveals that the initial implementation of the PART model, using all features, yields accuracy of 97.5162% but shows a high model build-up time of 61.23 seconds. Further refinement through the chi-square feature selection method identifies a subset of the top ranked 20 features based on chi-sqauare score, achieving an enhanced accuracy of 97.5202% with a significantly reduced model build-up time of 21.54 seconds. Comparative analysis with other feature selection methods establishes the superiority of the chi-square method for the Edge-IIoT dataset.