Hybrid Bayesian procedures for automatic detection of change-points

Hybrid Bayesian procedures for automatic segmentation of piecewise-constant univariate datasets are described. The procedures are designed to choose the number of discontinuities (referred to as change-points or break-points) present in the data as well as their positions. asic methods are introduced to reduce the size of the search space for otherwise computationally expensive Bayesian algorithms. In doing so, procedures retaining the benefits of Bayesian algorithms and having greatly reduced computation times are realised. These procedures may be used to give good results in cases where the full Bayesian analysis would be impossible due to the combinatorial complexity involved. Although no theoretical foundation is derived, simulation results show that this hybrid approach is a useful and practical one. ethods of reducing the search space are considered along with three Bayesian algorithms. The performance of the various schemes is investigated by application to a number of synthetic datasets and is compared with other segmentation algorithms.