Ensemble learning is considered as an important method to improve the accuracy of data mining and machine learning. On the base of the analysis of the basic concepts of ensemble learning, the design of ensemble learning model is divided into 3 stages: classifier construction, classifier integration, and classification result integration, then the method of increasing prediction accuracy were discussed from 3 aspects: controlling classifier error, enhancing generalization ability, and distinguishing acceptance-error in the application. Then, the influencing factors and the increasing methods of the 3 stages were studied through the experiments. The results show that it has great significance to reduce predication error, improving prediction accuracy, and construct a reasonable integrated learning model.