Soon M. Chung, Ph.D. (Advisor); Vincent A. Schmidt, Ph.D. (Committee Member); Nikolaos Bourbakis, Ph.D. (Committee Member)
Master of Science (MS)
With the rise of Big Data and the Internet of Things, there is an increasing availability of large volumes of real-time streaming data. Unusual occurrences in the underlying system will be reflected in these streams, but any human analysis will quickly become out of date. There is a need for automatic analysis of streaming data capable of identifying these anomalous behaviors as they occur, to give ample time to react. In order to handle many high-velocity data streams, detectors must minimize the processing requirements per value. In this thesis, we have developed a novel anomaly detection method which makes use of a diverse set of detectors in a hierarchical structure. The composite detector follows a filtration paradigm to mark each value in the series. The base model, chosen to be fast potentially at the expense of precision, identifies candidate anomalies in the series as each value arrives. Models higher in the hierarchy verify the candidates from their immediate predecessor, potentially rejecting some as false alarms. Our experiments show that this hierarchical method can achieve similar performance to state-of-the-art detectors using computationally simple models with lower processing requirements, enabling better scalability.
Year Degree Awarded
Copyright 2020, all rights reserved. My ETD will be available under the "Fair Use" terms of copyright law.