DIFT: A Diffusion-Transformer for Intrusion Detection of IoT with Imbalanced Learning

Class Imbalance Problem is a prevalent challenge in Internet of Things (IoT) intrusion detection, and deep learning models are prone to favor a larger number of classes leading to a severe degradation of the detection performance. To overcome the class imbalance problem and enhance the feature representation, we propose a novel IoT intrusion detection model named DIFT based on the Diffusion model and Transformer for imbalanced learning. Firstly, the Diffusion model is utilized to learn minority classes of sample feature patterns and generate a balanced dataset. Then, the IoT traffic data are locally feature-enhanced using the Patching method with shortened sequence length to improve processing efficiency, the local and global features of the balanced dataset are extracted using Time Series Transformer to improve the detection performance. Finally, we evaluate the effectiveness of the proposed method on the standard IoT intrusion detection datasets TON_IoT and DS2OS, the experimental results show that DIFT achieves superior performance compared with other remarkable methods, greatly improving the detection performance of imbalanced datasets.