Heart Disease Prediction Model Based on GBDT-RF

Heart disease affects the quality of life and even the safety of countless people. Exploring a model that can predict heart disease is significant for prevention and policy formulation. In this paper, a feature extraction model based on gradient boosting decision tree (GBDT) was established after preprocessing the data from the cardiac database of the University of California. The results showed that cholesterol level, thalassemia status and ST segment depression had the greatest contribution to the diagnosis of heart disease, and the characteristic contribution was $19.70 \%, 13.91 \%$ and $12.80 \%$, respectively. Subsequently, based on the gradient boosting decision tree (GBDT) features, the random forest (RF) heart disease prediction model is established, and four machine learning models, including BP neural network, random forest model, decision tree model and logistic regression model, are established for comparison. $A_{c c}=0.833, A R=0.820, A P=0.872$, F1-score $=0.845$, AUC $=0.862$, which indicates that the RF heart disease prediction model based on GBDT feature extraction can effectively predict the diagnosis of heart disease.