Method for detecting network intrusions utilizing VAE-CWGAN and combining statistical significance of features

Abstract

A detection approach based on VAE-CWGAN and the fusion of statistical relevance of features was presented in response to the issues with existing intrusion detection methods, which were limited by the class imbalance of datasets and the inadequate representation of selected features.To improve data quality, preparation of the data was done first.In order to address the issue of imbalanced datasets, a second VAE-CWGAN model was built to produce fresh samples, guaranteeing that the classification model was no longer skewed towards the majority class.In order to gather more representative features and improve the model's ability to learn data information, the features were then ranked using the standard deviation, difference between the median, and mean, as well as their statistical importance for feature selection.