Real-time segmenting time series data

Abstract

There has been increased interest in time series data mining recently. In some cases, approaches of real-time segmenting time series are necessary in time series similarity search and data mining, and this is the focus of this paper. A real-time iterative algorithm that is based on time series prediction is proposed in this paper. Proposed algorithm consists of three modular steps. (1) Modeling: the step identifies an autoregressive moving average (ARMA) model of dynamic processes from a time series data; (2) prediction: this step makes k steps ahead prediction based on the ARMA model of the process at a crisp time point. (3) Change-points detection: the step is what fits a piecewise segmented polynomial regressive model to the time series data to determine whether it contains a new change-point. Finally, high performance of the proposed algorithm is demonstrated by comparing with Guralnik-Srivastava algorithm. © Springer-Verlag Berlin Heidelberg 2003.