Thinning Random Forests

• Activity Number: 356
• Type: Contributed
• Date/Time: Tuesday, August 3, 2010 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistical Learning and Data Mining
• Author(s): Jie Xu* and J. Brian Gray+
• Companies: The University of Alabama and The University of Alabama
• Keywords: cluster analysis ; ensemble model ; machine learning ; predictive modeling

Abstract

Ensemble models, such as bagging, random forests, and boosting, have better predictive accuracy than single classifiers. However, an ensemble model typically consists of hundreds of single classifiers, which makes model interpretation difficult. In this presentation, we propose a new method based on cluster analysis for significantly reducing the size of an ensemble while maintaining roughly the same predictive accuracy. Model interpretation is improved, and computation time for future predictions is reduced. The method is demonstrated on several data sets and compared to other ensemble methods.