Time-series segmentation: A model and a method (Q1069634)

The problem of partitioning time series into segments is treated. The segments are considered as falling into classes. A different probability distribution is associated with each class of segment. Parametric families of distributions are considered, a set of parameter values being associated with each class. With each observation is associated an unobservable label, indicating from which class the observation arose. The label process is modeled as a Markov chain. Segmentation algorithms are obtained by applying a relaxation method to maximize the resulting likelihood function. Special attention is given to the situation in which the observations are conditionally independent, given the labels. A numerical example, segmentation of the U.S. gross national product, is given. Choice of the number of classes, using statistical model selection criteria, is illustrated.