Explainable Unsupervised Machine Learning for Cyber-Physical Systems

Cyber-Physical Systems (CPSs) play a critical role in our modern infrastructure due to their capability connect computing resources with physical systems. As such, topics such as reliability, performance, and security of CPSs continue receive increased attention from the research community. produce massive amounts data, creating opportunities use predictive Machine Learning (ML) models for performance monitoring optimization, preventive maintenance, threat detection. However, "black-box" nature complex ML is drawback when used safety-critical systems CPSs. While explainable has been an active area recent years, much work focused on supervised learning. rapidly unlabeled relying learning alone not sufficient data-driven decision making Therefore, if we are maximize CPSs, it necessary have unsupervised models. In this paper, outline how could be within We review existing ML, present initial desiderata CPS, Self-Organizing Maps based clustering methodology which generates global local explanations. evaluate fidelity generated explanations using feature perturbation techniques. The results show that proposed method identifies most important features responsible decision-making process Self-organizing Maps. Further, demonstrated strong candidate machine by comparing its model capabilities limitations current methods.