An Intelligent Machine Learning Framework for Enhancing IoT Security through Multi-Class Intrusion Detection

Abstract

The rapid expansion of Internet of Things (IoT) and Industrial Internet of Things (IIoT) environments has significantly increased the attack surface of modern cyber-physical infrastructures, making effective intrusion detection a critical security requirement. This study proposes a Unified Hybrid Artificial Intelligence-Driven Intrusion Detection System (UH-AIIDS) that integrates three benchmark datasets, namely Edge-IIoTset, CIC-IoT2023, and TII-SSRC-23, to enhance attack detection across heterogeneous environments. The framework combines advanced preprocessing, balanced sampling, ensemble machine learning models (CatBoost, LightGBM, and XGBoost), deep learning architectures, and a novel machine learning–deep learning decision fusion mechanism. Experimental evaluation demonstrates superior detection capability, robustness, and generalization performance. The proposed hybrid model achieved 88.22% accuracy on Edge-IIoTset and perfect binary classification performance on TII-SSRC-23, while effectively reducing inter-class confusion and improving cyber threat analytics. The framework provides a scalable and intelligent solution for next-generation IoT and IIoT cybersecurity.