Utilizing XAI Technique to Improve Autoencoder based Model for Computer Network Anomaly Detection with Shapley Additive Explanation(SHAP)

Machine learning (ML) and Deep Learning (DL) methods are being adopted rapidly, especially in computer network security, such as fraud detection, anomaly intrusion much more. However, the lack of transparency ML DL based models is a major obstacle to their implementation criticized due its black-box nature, even with tremendous results. Explainable Artificial Intelligence (XAI) promising area that can improve trustworthiness these by giving explanations interpreting output. If internal working understandable, then it further help performance. The objective this paper show how XAI be used interpret results model, autoencoder case. And, on interpretation, we improved performance for detection. kernel SHAP method, which shapley values, novel feature selection technique. This method identify only those features actually causing anomalous behaviour set attack/anomaly instances. Later, sets train validate autoencoderbut benign data only. Finally, built SHAP_Model outperformed other two proposed method. whole experiment conducted subset latest CICIDS2017 dataset. overall accuracy AUC 94% 0.969, respectively.