Product partition principal components model for multiple change point detection in multivariate data
Seminar Room 1, Newton Institute
The product partition model (PPM) is a well known method for multiple change point analysis but only limited to univariate data. In this article, an extension of PPM called product partition principal component analysis model (PPPCM) is proposed, which employs dimensionality reduction approach to the problem of multiple change point in multivariate data. The PPPCM can be used to detect distributional changes in the mean and covariance of a multivariate Gaussian data, providing a smaller dimensional representation of the data. The utility of the proposed PPPCM is demonstrated through experiments on simulated and real datasets. We also develop the model to address change-point analysis in "long" time series.