HIGH ACCURACY INTRUSION DETECTION IN IOT VIA HYBRID ML DL MODELS

Abstract

The rapid expansion of the Internet of Things (IoT) has introduced significant security vulnerabilities, making advanced Intrusion Detection Systems (IDS) a necessity. This research presents a high-accuracy hybrid framework that integrates Machine Learning (LightGBM) and Deep Learning (Artificial Neural Networks - ANN) for robust threat detection. Unlike traditional methods, the proposed system follows an anomaly-based detection approach to identify sophisticated cyber-attacks. The Light model is utilized for its high efficiency in classifying tabular network data, while the ANN component is designed to capture complex nonlinear patterns within the traffic. The framework was implemented and validated using the ACI-IoT-2023 dataset, which features a wide array of modern IoT attacks, including Port Scans, DDoS, and Brute Force. Experimental results demonstrate that this hybrid ML-DL architecture achieves exceptional detection accuracy and a significantly low false-positive rate, providing a scalable and effective security solution for heterogeneous IoT environments