An Efficient Anomaly Detection using Generative Adversarial Networks based on Minimax Loss Function

Recently, Artificial Intelligence (AI) is used as a vital tool for describing characteristics of information in the process of data mining and patterns. In data mining, an Anomaly Detection (AD) is an essential part which helps to determine behavior within the data and helps to detect network intrusion. The existing AD in network traffic has identified outliers in unbalanced datasets and difficult to train the model. Therefore, this paper introduces Generative Adversarial Networks (GAN) based on Minimax Loss Function for detecting anomalies in network traffic. Initially, this paper uses Neural Simulation Language-Knowledge Discovery in Databases (NSL-KDD) dataset which contains attacks and are preprocessed by Principal Component Analysis (PCA) to reduce the dimensionality. After that, these features are processed using the Information Gain (IG) algorithm for eliminating irrelevant features and then anomalies are detected by the proposed GAN-Minimax loss function which effectively identifies outliers, handles imbalanced data and trains the model effectively. The proposed GAN-Minimax resulted in high accuracy as 99.92% when compared with existing K-Nearest Neighbor (KNN) method.