In order to solve the problem of high intensity of image reading by doctors and easy misdiagnosis and missed diagnosis in early breast cancer screening, a classification model based on machine learning method was proposed for the benign and malignant diagnosis of mammography calcification images. After the CBIS-DDSM data set were preprocessed, the features of the region of interest image of the benign and malignant lesions on the mammography image of the patient were extracted. Lasso’s method was used to screen the 3 356-dimensional imaging omics features, and 74 features were obtained, which have the highest correlation with benign and malignant discrimination. Next, a variety of classification algorithms were combined with the classification model to perform its cross-validation training, predictive calculations were performed, and the receiver operating characteristic curve was used to evaluate the model. The results show that the model based on SMOTE-Lasso-RF method can get a better AUC and accuracy (including validation set AUC 0.812, ACC 0.938; test set AUC 0.736, ACC 0.739), which provides technical support for benign and malignant diagnosis of early breast cancer calcification mammography.