Abstract:The applications of random forests is used to predict commercial endowment insurance purchasing behavior .China's general social survey (CGSS) questionnaire survey data in 2017 is analyzed.SMOTE sampling is used to balance data set, then grid search is used to confirm mode input parameters .Finally the improved random forest model predictions is classified. And it is compared with support vector machine model. The results show that SMOTE oversampling method has a good performance in treating disequilibrium data, can improve the model performance, and the classification effect of stochastic forest after treatment is better than that of SVM.