Improved Interpretability with the use of Explainable AI for Intrusion Detection and Classification in Internet of Things Networks

Abstract

The exponential growth of the Internet of Things (IoT) has been phenomenal, changing the face of several sectors and making formerly mundane tasks much easier. There is a critical need to ensure the safety of IoT networks due to the fact that the expansion of IoT devices has created new entry points for cybercriminals. To protect the confidentiality and availability of IoT networks, intrusion
detection and classification systems are vital. Cyberattacks nowadays are very dynamic and complex, making it difficult for traditional
intrusion detection systems that rely on rules or signatures to stay up. Consequently, there has been a lot of focus on using machine
learning techniques for intrusion detection and classification in IoT networks recently. Intrusion detection systems can adapt and develop with new threats by using machine learning algorithms to automatically discover patterns and correlations from enormous amounts of data. The rapid expansion of IoT devices has increased security vulnerabilities, necessitating robust intrusion detection mechanisms. This research proposes a hybrid approach combining XGBoost for feature selection and deep learning models (CNN, LSTM, and MLP) for effective intrusion detection. We employ CNN for spatial feature extraction, LSTM for temporal pattern recognition, and MLP for classification. By examining data from network traffic, these algorithms may identify unusual behavior and group it into distinct types of attacks, safeguarding the system in real-time. Exploring and evaluating several machine-learning techniques for intrusion detection and classification in IoT networks is the goal of this project. The goal is to find the best models for detecting and correctly classifying intrusions in IoT settings. We will examine the efficacy of common machine learning methods including Decision Trees, Random Forest, Support Vector Machines (SVM), Neural Networks, and Ensemble Techniques by making use of the extensively used UNSW-NB15 dataset (Moustafa & Slay, 2016), which is an exhaustive dataset for network intrusion detection. We hope to find out what works and what doesn't by do an empirical evaluation of these algorithms using the UNSW-NB15 dataset. To evaluate how well the models, identify and categorize network intrusions, performance measures including recall, accuracy, precision, and F1-score will be utilized. In order to better protect IoT networks from cyber-attacks, this study will examine the possibility of using machine learning-based techniques.