A Method for Improving the Stability of Feature Selection Algorithm

This paper researches on problems of improving the stability of feature selection algorithm. A bagging-based selective results ensemble method is proposed. First use a feature selection algorithm and different training subsets to select several feature subsets. Then compute weights of each selected feature subset by mutual information and classifying accuracy. At last use a bagging-based method to assemble the selective subsets. Experiments in intrusion detection data of KDD cup´99 show that this algorithm could obtain better results.