Abstract

Info

Title:: Risk factors analysis and prediction of diabetic retinopathy based on Null Importance and GS-LGBM

Author(s):: CAO Jiayue; LUO Dongmei; School of Microelectronics and Data Science, Anhui University of Technology, Maanshan 243002, China

Keywords:: Keywords: diabetic retinopathy Null Importance risk profiling GS-LGBM

Abstract:: Abstract: Objective To analyze the risk factors of diabetic retinopathy (DR) and construct a DR risk prediction model through machine learning algorithms, thereby providing reference for DR prevention and diagnosis. Methods The study adopted the Diabetic Complication Early-Warning Data Set of the National Population Health Data Center. Null Importance method was used to remove noise features and screen out the key factors related to DR. LGBM model parameters were optimized with GridSearch to construct the GS-LGBM DR risk prediction model. The proposed method was compared with XGBoost, random forest, Logistic, and LGBM models in terms of accuracy, precision, recall, F1 score, and AUC values. Results Thirty key factors were screened out using the Null Importance method. Compared with XGBoost, random forest, Logistic and LGBM models, the GS-LGBM DR risk prediction model had the best evaluation performances, and its AUC value on the test data was as high as 0.897. Conclusion The hyperparameter optimized model is superior to the traditional DR prediction model, and it is more conducive to the clinical diagnosis of DR.