Purpose/Significance To construct a diabetic retinopathy risk prediction model integrating health profiles and machine learning，which realizes precise risk stratification and early lesion detection，and provides personalized health management for patients. Method/Process Based on a diabetic complication early warning dataset，the K-means method is used to establish health profiles. Sev. enteen machine learning models are adopted，which are modeled respectively based on the overall sample and each health profile. Tak. ing the overall optimal model as the baseline，the performance of the optimal models of each health profile is compared，and the inter. pretability of the models is analyzed through SHAP values. Result/Conclusion Five types of health profiles are identified. The optimal models outperform the overall baseline in terms of AUC and accuracy，indicating that the prediction model can improve the accuracy of risk identification for diabetic retinopathy.