Robust Deep Learning for Threat Classification and Intrusion Detection in Internet of Things Networks
Abstract
The fast growth of Internet of Things (IoT) network infrastructure has made advanced attacks more common, and therefore robust intrusion detection systems have become highly required. Traditional methods of detecting intrusions are faced with many problems when handling high-dimensional data that has a lot of noise and complicated interactions between features, which results in poor detection accuracy. In order to solve these problems, the new method of IoT Intrusion Detection and Threat Classification system based on an Explicit Feature Interaction-aware Graph Neural Network (IOT-IDTC-EFI-GNN) is suggested. First, the IoT-23 and BoT-IoT data sets are combined to enhance the variety of attacks and improve generalization. Next, the CNSAF technique is used to preprocess the data. In the next step, an explicit feature interaction-aware graph neural network (EFI-GNN) model is trained to learn intricate feature dependencies using graph-based messaging for efficient multiclass intrusion detection. According to the experimental results, the suggested method clearly outperforms current methods, achieving accuracy, precision, recall, F1 score, and specificity of 99.0%, 98.0%, 99.0%, 98.5%, and 99.0%, respectively. Thus, it can be said that the suggested method offers a precise and expandable solution for IoT network intrusion detection.