Traditional classification methods are based on the assumption that the training sets are well-balanced, however, in real case the data is usually unbalanced, and the classification performance of the traditional classification is always restricted. A detailed overview of domestic and foreign classification algorithms from the data level and algorithm level is provided in this paper. And through simulation experiments to compare the classification performance of a variety of unbalanced classification algorithm on six different data sets, it is found that the improved classification algorithm has varying degrees of improvement for overall performance. The paper concludes with a list of problems which need solving for the development of unbalanced data classification.