Enhancing IoT Botnet Detection through Machine Learning-based Feature Selection and Ensemble Models

Abstract

An increase in cyberattacks has coincided with the Internet of Things (IoT) expansion. When numerous systems are connected, more botnet attacks are possible. Because botnet attacks are constantly evolving to take advantage of security holes and weaknesses in internet traffic and IoT devices, they must be recognized. Voting ensemble (VE), Ada boost, K-Nearest Neighbour (KNN), and bootstrap aggregation are some methods used in this work for botnet detection. This study aims to first incorporate feature significance for enhanced efficacy, then estimate effectiveness in IoT botnet detection using traditional model-based machine learning, and finally evaluate the outcomes using ensemble models. It has been demonstrated that applying feature importance increases the effectiveness of ensemble models. VE algorithm provides the best botnet traffic detection compared to all currently used approaches.