A new proposal to speed up one-class projectionist techniques

Abstract With the increasing complexity of cyber-physical systems, combined with the potential risk of cyber-attacks, the possibility of having reliable anomaly detection systems plays a significant role in every process, promoting the scientific community’s efforts. Due to the remarkable importance of this kind of tool, a wide variety of approaches and techniques model the expected behavior of a system. An intuitive way to deal with this problem consists of determining the boundaries of the dataset. This document presents a novel procedure to speed up an existing anomaly detection technique based on non-convex boundaries.