Abstract: Ensemble learning is an important research content of machine learning. By integrating or combining existing machine learning models, ensemble learning can make the performance of the integrated model exceed any single model. The theory of the effectiveness of ensemble learning is summarized and analyzed from two aspects: ensemble regression and ensemble classification; the methods to improve the diversity of ensemble learning are analyzed; the research progress of the main ensemble learning methods such as bagging, boosting, stacking, multi-kernel learning and ensemble deep learning is analyzed; the key issues of ensemble learning to be focused in the future are discussed.