Traditional Belief-Rule-Based Ensemble learning methods usually integrate all sub-BRB systems that are trained to obtain better results than a single belief-rule-based system.As the number of BRB systems participating in ensemble learning increases and a large amount of redundant sub-BRB systems are generated because of the reducing of the difference between subsystems,which drastically result in the decreasing prediction speed and the increasing required storage for the BRB systems.In order to solve the above problems,this paper proposed a selective ensemble learning approach for the BRB classification system(BRBCS)base on the Pareto Archived Evolutionary Strategy(PAES)multi-objective optimization,which employed the improved Bagging algorithm to train the base classifier.With the purpose of increasing the degree of difference in the integration of the base classifier,the training set was constructed by repeated sampling of data.In the base classifier selection stage,the base classifiers participating in the integration were binary coding; then the number of base classifiers participating in the integration and the generalization error of the classifier was conducted as the objective function in the above multi-objective optimization problem and finally the elite retention strategy and the adaptive mesh algorithm were adopted to solve PAES optimal solution set.In order to verify the effectiveness of this method,three case studies on classification problems were performed to illustrate how the efficiency of the BRBCS-PAES method.The Comparison results demonstrate that the proposed method can effectively reduce the number of base classifiers participating in the integration and improve the accuracy of BRBCS.