In recent years, imbalanced learning has attracted the attention of many researchers. In general, minority classes are more noteworthy, and the cost of misclassification is much higher than that of majority classes. Because of the imbalanced distribution of imbalanced data, the standard classification algorithms will be difficult to apply. In order to solve the problem of imbalanced data classification, a zeroth-order optimization algorithm based on under-sampling is presented. Firstly, in order to reduce the influence of imbalanced data distribution, two different sampling strategies are adopted for data sets with different imbalanced ratios. Then, an SVM(Support vector machine) model with margin mean term is used for classification, and a zeroth-order stochastic gradient descent algorithm with reduced variance is used to solve the problem. At the same time, the accuracy of the algorithm is improved. A comparative experiment is carried out on imbalanced data, and the experimental results show that the proposed method effectively improves the classification effect of imbalanced data.