二型糖尿病肾病风险预测模型的比较

目的：选择相应的机器学习算法构建二型糖尿病肾病风险预测模型，为疾病的早期预防提供科学依据。方法：基于解放军总医院提供的糖尿病数据集，通过对缺失值、异常值等进行一系列预处理，得到894条二型糖尿病患者数据。利用单因素逻辑回归筛选出24个有效检查指标作为特征，并基于随机森林、BP神经网络、支持向量机分别构建二型糖尿病肾病风险预测模型，同时对其查准率、召回率进行对比，以验证其应用性能。结果：随机森林预测模型的总体性能最优，3种算法的训练效果均较好。结论：二型糖尿病肾病风险预测模型能为疾病早期预防控制提供参考依据。

Risk prediction models of type 2 diabetic nephropathy

Objective To provide the scientific evidence for the risk prediction models of type 2 diabetic nephropathy established by different machine learning algorithms for the early prevention of diseases. Methods Eight hundred and ninety-four data of type 2 diabetic patients were obtained by preprocessing the missing and abnormal values based on the diabetic data set provided by the Chinese PLA General Hospital. Twenty-four effective examination indicators screened from the 894 data by univariate logistic regression analysis were used as the characteristic indicators. Three risk prediction models of type 2 diabetic nephropathy were established using the random forest algorithm, BP neural network algorithm and support vector machine algorithm respectively. Their precision ratio and recall ratio were compared to verify their applicability. Results The overall performance of the risk prediction model of type 2 diabetic nephropathy established by random forest algorithm was the better than that of those established by BP neural network algorithm and support vector machine algorithm. The training effect of the three algorithms was good. Conclusion The three risk prediction models of type 2 diabetic nephropathy can provide reference and evidence for the early prevention of diseases.