Time series clustering with Dirichlet process mixtures of ARMA models

点击翻译按钮获取中文摘要

ARMA models are parsimonious stochastic models for time series. Given a set of time series, we can cluster them by regarding that they were drawn from a mixture of ARMA models and by fitting the model to them, where if the mixture model is a Dirichlet process mixture, the number of the clusters can be simultaneously estimated. In this work, we have developed a Markov Chain Monte Carlo method to fit a Dirichlet process mixture of ARMA models to a set of time series.