Adversarial Anomaly Detection Using Centroid-Based Clustering

摘要: As cyber attacks are growing with an unprecedented rate in the recent years, organizations seeking efficient and scalable solution towards a holistic protection system. adversaries becoming more skilled organized, traditional rule based detection systems have been proved to be quite ineffective against continuously evolving attacks. Consequently, security researchers focusing on applying machine learning techniques big data analytics defend Over several anomaly claimed successful sophisticated including previously unseen zero-day But often, these do not consider adversary's adaptive attacking behavior for bypassing procedure. result, deploying active real-world scenarios fails provide significant benefits presence of intelligent that carefully manipulating attack vectors. In this work, we analyze adversarial impact models built upon centroid-based clustering from game-theoretic aspect propose technique models. The experimental results show our can withstand effectively compared