基于Null Importance与GS-LGBM的糖尿病视网膜病变因素分析与风险预测

Risk factors analysis and prediction of diabetic retinopathy based on Null Importance and GSLGBM

目的:通过机器学习算法分析糖尿病视网膜病变(DR)关键因素,构建DR风险预测模型,为DR的预防和诊断提供参考。方法:采用国家人口健康科学数据中心的《糖尿病并发症预警数据集》,基于Null Importance方法去除噪声特征,筛选出与DR有关的关键因素;基于GridSearch优化LGBM模型参数,构建GS-LGBM DR风险预测模型。以准确率、精确率、召回率、F1分数、AUC值作为评价标准,与XGBoost、随机森林、Logistic以及未调优的LGBM模型进行比较。结果:Null Importance方法筛选出30个关键因素;与XGBoost、随机森林、Logistic以及未调优的LGBM模型相比,本研究所构建的GS-LGBM DR风险预测模型各评价指标均最优,其在测试数据上的AUC值高达0.897。结论:相较传统的DR预测模型,经过超参数优化后的模型具有更好的DR风险预测能力,更有助于DR的临床诊断。

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.