Hierarchical Bagging Ensemble Pruning Based on the Diversity of Base Learners

The main objective of this paper is to find a rapid pruning method for Bagging to reduce the storage space needed by the algorithm,speed up the computation process and obtain the potential of improving the classification accuracy.A new idea of selective ensemble is proposed,which computes the diversity of base learners directly.The base learner which has the strongest ability to improve the diversity of other base learners in the base learner set is chosen and deleted,and hierarchical pruning is used to speed up the new algorithm.The new algorithm can greatly reduce the size of the bagging ensemble without performance degradation.It also supports parallel computing and its selective ensemble speed is much faster than that of GASEN(genetic algorithm based on selected ensemble).The upper bound of classification error of ensemble learning is given.