Optimization of Time-series Data Partitioning for Anomaly Detection

The concepts of symbolic dynamics and data set partitioning have been used for feature extraction and anomaly detection in time series data. Although modeling of state machines from symbol sequences has been widely reported, similar efforts have not been expended to investigate partitioning of time series data to optimally generate symbol sequences for anomaly detection. This paper addresses this issue and proposes a partitioning method based on maximum migration of data points across cell boundaries. Various aspects of the proposed partitioning tool, such as adaptiveness of alphabet size selection, noise mitigation, and robustness to spurious disturbances, are discussed. Experimental results on laboratory apparatuses of electronic circuits and electric motors show that maximum-migration partitioning yields significant improvement over existing partitioning methods (e.g., maximum entropy partitioning) for the purpose of anomaly detection.