Pruning of Random Forest classifiers: A survey and future directions

Abstract

Random Forest is an ensemble supervised machine learning technique. Based on bagging and random feature selection, number of decision trees (base classifiers) is generated and majority voting is taken for classification. For effective learning and classification of Random Forest, there is need for reducing number of trees (Pruning) in Random Forest. We have presented here systematic survey of pruning efforts of Random Forest classifier along with the required theoretical background. Most of the work for pruning takes static approach while recently dynamic pruning is being targeted. We have also generated a Comparison Chart by taking relevant parameters. There is research scope for analyzing behavior of Random forest, generating accurate and diverse base decision trees, truly dynamic pruning algorithm for Random Forest classifier, and generating optimal subset of Random forest.