A Deep Learning-Based Approach with Overlapped Classes Aggregation for Intrusion Detection in Internet of Things Networks
Salah Eddine Maoudj, Aissam Belghiat · 2024
The rise of Internet of Things (IoT) technology has enhanced several aspects of our lives. However, the dynamic connection between IoT devices, the resource restrictions, and their heterogeneity elevate the risk of network breaches. The intrusion detection system (IDS) is an efficient tool for network traffic investigation. Nevertheless, besides the imbalanced nature of cybersecurity datasets, some also suffer from the class overlap problem. The presence of one of these issues can limit the full potential of the IDS. However, their joining in a dataset drastically degrades its performance when adopting the multi-classification case. Therefore, novel techniques should be explored to handle the attendance of class imbalance with class overlap in cybersecurity datasets. In this paper, we propose a deep learning-based approach with overlapped classes aggregation for intrusion detection enhancement in IoT networks. Each group of overlapped classes is treated as a single class. By doing so, we have eliminated the class overlap and mitigated the class imbalance, resulting in an improved overall performance in exchange for less specification on the type of attack that intruded on the network. In addition, we tackled the remaining imbalance utilizing the class weight technique. The proposed approach surpasses state-of-the-art methods by achieving 88.96% accuracy and high individual detection rates for all the classes on the UNSW-NB15 dataset.