AWDLID: Augmented WGAN and Deep Learning for Improved Intrusion Detection

The efficacy of machine learning models in terms of accuracy and speed poses a significant challenge to AI-driven intrusion detection. This work introduces an AI-driven approach, termed AWDLID, which improves the quality of the training dataset and utilizes deep learning models to enhance intrusion detection efficacy. AWDLID comprises AWGAN and a deep learning model referred to as DLID. AWGAN is an innovative method that enhances dataset quality by integrating Wasserstein GAN with K-means algorithms. Consequently, we use WGAN to generate more realistic samples from minority classes and K-means to find appropriate samples from majority classes. Moreover, we create the deep neural network DLID for intrusion classification to analyze and predict traffic flows. The comprehensive experiments utilizing established datasets, such as NSL-KDD and CSE-CIC-IDS2018, enable us to validate the effectiveness of our proposed AWDLID. The results show that AWDLID show superior performance to state-of-the-art methods with an outstanding F1-score of 99.21% and 98.69%.