The classification result of classical support vector machine(SVM) algorithm in the case of unbalanced data set is not satisfactory. In order to improve the SVM algorithm’s classification performance under unbalanced data set, a novel under-sampling algorithm based on optimization of decreasing reduction(ODR) is presented. The algorithm is applied to under-sample the majority class instances for removal of a large number of overlapping samples of redundant and noise samples, which consequently makes reservations for the majority class instances with more useful information, and the ODR under-sampling algorithm is combined with border synthetic minority over-sample technique(BSMOTE) to achieve a balanced training sample data set. The experimental results show that the proposed method can not only improve classification performance of SVM in the minority class data, but also increase the overall classification performance.a