ML/CGAN: Network Attack Analysis Using CGAN as Meta-Learning

Abstract

This letter proposes a deep learning (DL) technique that uses Conditional GAN (CGAN) as meta-learning to overcome the network traffic data shortage in attack analysis. The technique consists of four parts: 1) Implement CGAN and MLP regression in IDS/Traffic classifier. 2) Preprocess network traffic data and divide it into training and test sets. 3) Generate data including labels by the trained CGAN and concatenate it to the training set. 4) Train MLP regression using the expanded dataset. The experiment is executed with the traffics provided by CAIDA and CIDDS-001. Numerical results show that the proposed technique enables us to overcome data shortage and outperforms MLP regression in measurements used to analyze network traffic. This study also suggests several discussions for network traffic analysis using GANs, such as the relationship between GAN generator loss and its performance.