A new algorithm for segmenting data from time series

This paper addresses the problem of dividing a time series into segments, where the data in each segment, is generated by a different underlying linear model. The technique is used to identify changes in the process generating the data. The identification of changes is recast as a shortest path problem which is solved using dynamic programming. The algorithm determines the total number of jumps within the data, the location of these jumps and the order of the model within each segment. Results of the application of the algorithm to the analysis of retail sales data are presented.