Ensemble Learning Approach for Robust Cyberattack Detection in Heterogeneous IoT Networks

Abstract

Although rapid development of IoT (Internet of Things), the security issues are becoming serious in recent years because of its heterogeneous traffic with diverse behaviors and highly dynamic traffic pattern. This paper presents a robust and adaptive Weighted Ensemble Learning (WELL) framework to detect the intrusion of IoT networks, which combines systematically data preprocessing, correlation, based feature selection, ensemble of complementary multiple classifiers (RF, XGBoost, SVM and KNN) based weighted voting, and is evaluated by the contemporary resilient CICIoT2023 data, set containing different IoT traffic types and various attacks (DoS, DDoS, Malware, Scanning etc.). It is demonstrated that our ensemble model successfully exceeded all individual classifiers in terms of accuracy of 98.1% and F1 score of 0.981, while still kept an extremely low false alarm rate of 1.6%. It can detect complicated attack from heterogeneous traffic with low false positive rate and adaptively adapt to dynamic traffic patterns. This trusted and scalable reliable system can promise in real time application and shows great potential of ensemble method to boost reliability and efficiency in IoT networks.