In this paper, a new method was proposed to reduce the misclassification cost of software defect prediction under the condition of imbalanced classes. The effectiveness of the method was evaluated by the correct rate of sample classification, precision rate, recall rate, and F-Measure index. The main results of this research were as follows: (1) The proposed method can maintain a high accuracy rate while maintaining a relatively low cost of misclassification; (2) In the data preprocessing stage, a median assignment method is proposed, Used to deal with the field missing value problem of a reasonable sample in the data set; (3) In the classification stage of the decision tree and the voting classification stage of the formed random forest, the cost-sensitive factors defined according to different objects are introduced respectively, and the training is based on the cost-sensitive The improved random forest model. Experimental results show that this method can reduce the cost of misclassification while maintaining a high accuracy rate.