Hybrid Feature Selection Framework for Building Resource Efficient Intrusion Detection Systems Model in the Internet of Things

Abstract

The Internet of Things (IoT) infrastructure exhibits numerous security vulnerabilities which have led to an increasing number of cyber-attacks. This problem has led to the development of Intrusion Detection Systems (IDSs) explicitly tailored for detecting cyber-attacks on IoT infrastructure, particularly IDSs that utilize machine learning models. To achieve high-performance machine learning models, the training data is pivotal. However, IoT infrastructure has limited computational resources, especially CPU, memory, and power. One solution is reducing the amount of data by selecting only certain features that have a significant impact on the quality of the machine learning models. This study proposes a novel hybrid framework to select the important features to build an efficient IDS model for IoT using the XGBoost algorithm. The performance of the proposed framework was rigorously tested using the network TON-IoT dataset, which is specifically designed to develop IDSs for IoT infrastructure. The experimental results show that our proposed framework yields machine-learning models with optimal performance while conserving computational resources. Through the implementation of our framework, we achieve a substantial reduction in features by 78.95%, coupled with a 3.18% Macro F1-score improvement. Furthermore, the framework also yields significant reductions in training, validation, and testing durations by 53.25%, 41.94%, and 47.06%, respectively. Also, it yields a model with impressive reductions in memory usage by 81.82%.