Title: Combining structural change analysis with anomaly detection: A constrained clustering approach
Authors: Carlo Drago - University of Rome Niccolo Cusano (Italy) [presenting]
Abstract: The identification of the structural changes is a very important problem in modern time series analysis. We integrate the structural change analysis with the anomaly detection also defined as the specific identification of the deviations from a given pattern. It is proposed an approach for decomposing the time series in components and identifying the different structural breaks of the time series. In this sense it is considered a constrained hierarchical clustering algorithm in order to detect the anomalies and the structural changes over time. The approach is relevant in order to simultaneously determine the structure of the time series, the structural breaks and the different anomalies (single or subsequent observations) over time. The anomalies can occur considering the direction and the relevance of the structural changes over time. The approach is investigated by using simulated and also real time series.