Change point detection in piecewise stationary time series for animal behavior analysis

Detection of abrupt changes in time series data structure is very useful in modeling and prediction in many application areas, where time series pattern recognition must be implemented. Despite of the wide amount of research in this area, the proposed methods require usually a long execution time and do not provide the possibility to estimate the real changes in variance and autocorrelation at certain points. Hence they cannot be efficiently applied to the large time series where only the change points with constraints must be detected. In the framework of the present paper we provide heuristic methods based on the moving variance ratio and moving median difference for identification of change points. The methods were applied for behavior analysis of farm animals using the data sets of accelerations obtained by means of the radio frequency identification (RFID).