IGAN: Intrusion Detection Using Anomaly-Based Generative Adversarial Network

We present an architecture, an Anomaly-based Generative Adversarial Network (IGAN), that detects malicious strings with decent accuracy. The proposed IGAN is comprised of an encoder, a decoder, and a discriminator. The encoder and decoder form the generative unit, trying to reconstruct the input and map that input and output to a latent space variable. IGAN exploits this latent space together with the adversarial training with the discriminator to enhance the learning of the normal distribution. The discriminator used is as a classifier as well as a feature extractor. To identify an anomaly or a normal instance, the anomaly score is calculated after passing it to the trained model. We have performed a detailed analysis of the existing and the proposed architecture using reliable metrics such as AUC score, Precision, Recall, and F_score. The experimental results of the proposed IGAN outperform other existing models in detecting anomalies with high accuracy.