针对支持向量机在成绩预测时面临的泛化能力不足问题,提出基于最小流形类内离散度支持向量机M2 SVM。实验选取2000-2009年59次刘翔110m栏成绩作为研究对象,首先将前54次成绩作为训练样本并对模型进行训练得到分类标准,然后将后5次成绩作为测试样本并依次输入模型,比较预测结果与实际结果之间的相似程度,从而说明所提方法的有效性。该方法对人才选拔、成绩提升和梯队建设等具有重要意义。 In view of the lack of generalization ability of SVM in performance prediction, this paper proposes a M2 SVM based on minimal manifold support vector machine (SVM). The experiment selected Liu Xiang 110m column from 59 to 59 times from 2000 to 2009 as the research object. Firstly, the first 54 scores were taken as the training samples and the models were trained to get the classification criteria. Then the last 5 scores were used as the test samples and then input into the model. The degree of similarity between the results and the actual results shows the effectiveness of the proposed method. This method is of great significance to talent selection, achievement improvement and echelon construction.