A New AECM Algorithm Tuned to a Faster Beat for Model-Based Clustering of Regression Time-Series Data

• Activity Number: 63
• Type: Topic Contributed
• Date/Time: Sunday, July 31, 2011 : 4:00 PM to 5:50 PM
• Sponsor: Section on Statistical Learning and Data Mining
• Author(s): Wei-Chen Chen*+ and Ranjan Maitra
• Companies: Iowa State University and Iowa State University
• Keywords: statistical learning ; mutual funds ; EM algorithm ; finite mixture model ; model-based clustering ; autoregressive model

Abstract

We propose a model-based clustering time series regression model in an unsupervised machine learning framework to identify groups in regression time series data, where it is assumed that the mixture components follow an Gaussian autoregressive regression model of order p. Given the number of groups, the model has a mixture structure and can be maximized by the EM algorithm. With extra data augmentation scheme, we propose an alternative partial expectation conditional maximization algorithm (APECM), to improve the slow convergence of the EM algorithm, and show improved performance in both numbers of iterations and computing time. The methodology is applied to the context of clustering mutual funds data on the basis of their rates of return in the presence of economic indicators.