Heart disease, as one of the most serious vascular diseases that threaten human life and health in today's society, not only seriously threatens human life safety, but also brings serious economic burden to families and society due to high treatment costs. Aiming at the problems of insufficient accuracy and lack of feature interpretability in the current heart disease prediction research, by mining the important features that affect heart disease, the accurate prediction of heart disease and the interpretability analysis of influencing factors are realized. First, use the T test to analyze the significant difference (P-value) between the features, and select the features to combine by descending the P-value value. Then, the prediction of heart disease and its feature interpretability analysis were implemented using ten machine learning models and SHAP methods. Validation experiments were performed on the UCI heart disease dataset, and it reached 1 on seven evaluation indicators widely used in medical fields, which was better than and compared with the experimental results of the paper. Finally, the SHAP method is used to analyze the interpretability of 13 features, and the results are visualized through feature importance ranking, and the association between a single feature and heart disease can be mined, which can provide decision support for doctors in precision medicine for heart disease.