Belief Rule-Based (BRB) model has been widely used for complex system classification and prediction. However, excessive antecedent attributes will cause the combinatorial explosion problem, which restricts the applicability of the BRB model to high-dimensional problems. In this paper, we propose an Ensemble-BRB model with the use of the bagging framework to downsize the belief rule base and avoid the combinatorial explosion problem. The kernel of the Ensemble-BRB model is to generate several weak BRBs orderly, each of which only consists of a subset of antecedent attributes. Different combination methods can be used to integrate these weak BRBs coherently for classification and prediction respectively. Four benchmark problems are tested to validate the efficiency of the proposed Ensemble-BRB model in classification, and a real case on the health index prediction of engines proves the feasibility of the Ensemble-BRB model in prediction. The results on both classification and prediction show that the Ensemble-BRB model can effectively downsize the BRB as well as reach a high modeling accuracy.