To explore the key influencing factors of diabetic retinopathy (DR), analyze the current situation of DR, and construct a risk prediction model.

Based on the diabetic complication early warning data set published by the national population and health science data sharing platform, the key influencing factors of DR were obtained by univariate and multivariate Logistic regression analyses. The entropy weighting method, the Technique for Order Preference by Similarity to Ideal Solution (TOPSIS), and the rank-sum ratio (RSR) were used to quantify the risk of DR development in patients and stratified into three levels: high, medium, and low. Logistic regression, random forest, and support vector machine models were constructed, respectively, and model fusion was performed using voting, averaging, and weighted averaging to evaluate the model predictive effect and obtain the best predictive model.

Finally, 14 indexes including age and hyperlipidemia were extracted as key influencing factors. The stratification results showed that there were 50 diabetic patients without DR in this data set with a risk of about 82.99%, which was a high-risk group for DR and needed more attention. The best prediction effect was obtained from the voting machine fusion model (Acc: 80.18%, F1: 0.7868).

The key influencing factors of DR are analyzed, providing the direction of treatment and prevention. The low, medium, and high-risk groups of DR are classified for risk early warning. By comparing the effect between the models, the prediction model of morbidity risk of DR is constructed, providing insights for clinical early warning and data analysis.