DID-IDS: A Novel Diffusion-based Imbalanced Data Intrusion Detection System

With the continuous expansion of the volume of Internet, the methods and forms of network attacks have also improved and developed, which requires advanced intrusion detection system. Although deep learning methods achieve remarkable success in the accuracy of intrusion detection over traditional methods, a great number of balanced data are required in the training process. However, in the real-life circumstance, the ratio between benign and abnormal data is not balanced. To address this issue, this paper proposes a novel intrusion detection system, called Diffusion-based Imbalanced Data Intrusion Detection System (DID-IDS). In this system, in order to enhance the correlation between data features, the table data are firstly transformed into the image data. Then, the converted images are sent to the diffusion model to generate high-quality fake images for data enhancement purposes. Finally, the enhanced data set is used to train the Convolutional Neural Network (CNN) model for intrusion detection to improve prediction accuracy. The experimental results show that the proposed system has better performance compared to intrusion detection systems using traditional or Generative Adversarial Networks (GAN) based data augmentation methods.