XA-GANomaly: An Explainable Adaptive Semi-supervised Learning Method for Intrusion Detection using GANomaly in Global Economic dynamic shifts

Abstract

Economic Intrusion detection is discovering unauthorized net-work activity and recognizing whether the data is an abnormal network transmission. Recent research has focused on semi-supervised learning mechanisms for identifying abnormal net-work traffic to deal with both labeled and unlabeled in industry. Economic shift generation is very crucial in creating a balance between hyper sensitive generation and superficial balance. However, real-time training and classifying network traffic remains as challenges, as they can lead to the degradation of the overall dataset and difficulties in preventing attacks. Addi-tionally, existing semi-supervised learning research may not comprehensively analyze the experimental results. To address these issues, we propose XA-GANomaly, a novel technique for explainable adaptive semi-supervised learning using GANoma-ly, an image anomalous detection model, that sequentially trains small subsets in a dynamic way. First, we introduce a Deep Neural Network (DNN) based GANomaly for semi-supervised learning. Second, we present our proposed adaptive algorithm for DNN-based GANomaly, validated with four sub-sets of the adaptive dataset. Finally, we demonstrate a monitor-ing system that incorporating three explainable techniques – SHapley Additive exPlanations (SHAP), reconstruction error visualization, and t-distributed stochastic neighbor embedding (t-SNE) – to respond effectively to attacks toward traffic data at each stage of feature engineering, semi-supervised learning, and adaptive learning. Compared to other single-class classifi-cation techniques, the proposed DNN-based GANomaly achieves higher scores than 13% and 8% of F1 scores and accuracy of 4.17% and 11.51% for the NSL-KDD and UNSW-NB15 datasets, respectively. In addition, our adaptive learning experiments show mostly improved results over initial values, and we provide an analysis and monitoring system based on the combination of the three explainable methodologies. As a result, our proposed method has the potential to be applied in the real industry, and future research will explore on handling unbalanced real-time datasets in various scenarios.The essence of such shifts in economic paradigm can be better understood by applying various triggers in functional machine to machine interaction.This can be better understood by creating a super-vised paradigm which better understands global economic shifts.