Data-Driven Anomaly Detection based on a Bias Change
2014 (English)In: Proceedings of the 19th IFAC World Congress, 2014Conference paper (Refereed)
This paper proposes batch and sequential data-driven approaches to anomaly detection based on generalized likelihood ratio tests for a bias change. The procedure is divided into two steps. Assuming availability of a nominal dataset, a nonparametric density estimate is obtained in the first step, prior to the test. Second, the unknown bias change is estimated from test data. Based on the expectation maximization (EM) algorithm, batch and sequential maximum likelihood estimators of the bias change are derived for the case where the densit yestimate is given by a Gaussian mixture. Approximate asymptotic expressions for the probabilities of error are suggested based on available results. Simulations and real world experiments illustrate the approach.
Place, publisher, year, edition, pages
IdentifiersURN: urn:nbn:se:liu:diva-109334OAI: oai:DiVA.org:liu-109334DiVA: diva2:737667
19th IFAC World Congress, Cape Town, South Africa, August 24-29, 2014