Feature Selection Methods for IoT Intrusion Detection System: Comparative Study

Abstract

With the advancement in the Internet of Things (IoT), more applications and services are deployed in information system and physical space. A huge volume of data is gathered from these devices through sensors embedded in the IoT environment. This massive volume of data can be exploited by malicious users and is vulnerable to various types of attacks. Therefore, various intrusion detection systems (IDS) for the IoT infrastructure have been developed to ensure the security of these devices. For improving the performance of IDS, feature selection (FS) methods play a crucial role. Feature selection methods reduce dimension of data and help machine learning (ML) classifiers in achieving better accuracy. In this paper, performance of four FS methods is compared for IoT-based IDS. The four bio-inspired FS methods used are whale optimization (WO), salp swarm algorithm (SSA), Harris hawk optimization (HHO), and gray wolf optimization (GWO). The classification of intrusive traffic is performed by two ML classifiers: K-nearest neighbor (KNN), and Naïve Bayes (NB). Experimental results show that GWO with KNN classifier outperforms other FS methods with the highest accuracy, precision, recall, F-score, and takes less execution time. The BoT-IoT dataset is used for system performance evaluation.