A two-level detector of short-term unique changes in time series based on a similarity method

Abstract

In the paper, a novel two-level algorithm of time-series change detection is presented. In the first level, to identify non-stationary sequences in a processed signal, preliminary detection of events is performed with a short-term prediction comparison. In the second stage, to confirm the changes detected in the first level, a similarity method aimed at identification of unique changes is employed. The detection of changes in a non-stationary time series is discussed, implemented algorithms are described and the results produced on a sample four financial time series are shown. General conditions for implementing the proposed algorithm as an immune-like event detector are discussed.