Abstract: The random forest algorithm is suitable to deal with high dimensional data sets with high correlative features, and its variable importance measure provides a natural method for feature selection. A feature selection algorithm based on random forest (RFFS) was proposed. RFFS adopted random forest algorithm as the basic tool, the classification accuracy as the criterion function, and the sequential backward selection and generalized sequential backward selection for feature selection. The experimental results on UCI datasets show that RFFS algorithm has better performance in classification accuracy and feature selection subset than the other methods in literatures.

Key words: artificial intelligence, random forest, feature selection, wrapper