Research on risk prediction model of retinopathy in patients with type 2 diabetes based on machine learning
LI Linyi
WAN Yi
LIU Kui
HAN Zhengyuan
HAN Tixin
CHEN Changsheng
Abstract：Objective To analyze and compare the predictive performance of six machine learning models,namely logistic regression(LR),linear discriminant analysis(LDA),decision tree(DT),random forest(RF),CatBoost,and LightGBM,in predicting the risk of retinopathy in patients with type 2 diabetes.Methods Based on the early warning dataset of diabetic complications from the National Population Health Data Center,retinopathy risk prediction models for type 2 diabetic patients were constructed based on different algorithms.After data cleaning,the training dataset and test dataset were divided according to 7∶3,and the prediction performance(accuracy,recall,precision,F1-Score,and AUC)of the six models was compared using the ten-fold cross-validation method.Results In the test dataset,six models were used to predict the risk of retinopathy in patients with type 2 diabetes,and LightGBM exhibited the highest accuracy(0.773 8),recall(0.825 1),precision(0.762 0),F1-Score(0.792 3),and AUC(0.843 4).CatBoost followed with accuracy(0.744 1),recall(0.787 8),precision(0.729 1),F1-Score(0.757 3),and AUC(0.828 0).Conclusion For predicting the risk of retinopathy in patients with type 2 diabetes,LightGBM performs the best among the six models.Additionally,ensemble learning classifiers(LightGBM,CatBoost,and RF)outperform weak learning classifiers(DT,LR,and LDA)in terms of predictive performance,offering a convenient and reliable reference for clinical auxiliary diagnosis and risk prediction of retinal complications in patients with type 2 diabetes.
Keywords：diabetic retinopathymachine learningdisease predictioncomparative study
Publication Date：2025-08-31
Online Publishing Date：2026-08-26(First online date of this platform, not the publication date of the document)
Pages：6( 1070-1075 )