Candidate Model Choice in Feature-Based Model Combination

19 Pages Posted: 17 Jan 2011

See all articles by Mingyang Xu

Mingyang Xu

American International Group, Inc.

Michael Golay

affiliation not provided to SSRN

Date Written: March 1, 2007

Abstract

This paper is intended to solve a central problem in recently developed feature-based model combination method (Xu and Golay, 2005), that is, how to choose candidate models. Through our analysis, we first conclude that the efficiency of model combination highly depends on the choice of candidate models. Some desirable properties are then proposed to assess a group of candidate models, which include accuracy, diversity, independence as well as completeness. To facilitate the choice with the use of these criteria, some quantitative measures are put forward. Meanwhile, Bayesian method and utility function are employed to aggregate information to obtain an overall evaluation of models. Finally, a stepwise forward candidate model choice procedure is proposed to realize all these criteria in a procedure, which chooses a group of candidate models out of a model pool.

Keywords: Candidate model choice, feature-based model combination, model assessment, Bayesian information aggregation

JEL Classification: C13, C14, C44, C51, C52, C61

Suggested Citation

Xu, Mingyang and Golay, Michael, Candidate Model Choice in Feature-Based Model Combination (March 1, 2007). Available at SSRN: https://ssrn.com/abstract=1742032 or http://dx.doi.org/10.2139/ssrn.1742032

Mingyang Xu (Contact Author)

American International Group, Inc. ( email )

80 Pine Street
New York, NY 10270
United States

Michael Golay

affiliation not provided to SSRN ( email )

Do you have negative results from your research you’d like to share?

Paper statistics

Downloads
47
Abstract Views
707
PlumX Metrics