2型糖尿病肾病预测的集成学习模型研究

Ensemble learning model for predicting type 2 diabetic nephropathy

目的:构建3个2型糖尿病肾病的预测模型。方法:基于国家临床医学科学数据中心的《糖尿病并发症预警数据集》,采用多种预处理方法清洗数据,通过Logistic回归分析和互信息等方法筛选特征。选用提升决策树、决策森林和决策丛林3种集成学习模型进行训练,实现对2型糖尿病肾病的风险预测。采用准确率、精确率、召回率,F1值和AUC 5个指标评价和比较3种模型的预测效能。结果:3个基于集成学习的糖尿病肾病预测模型的预测水平较高,其中二分类提升决策树模型的预测性能最优。结论:基于集成学习的糖尿病肾病预测模型可以为2型糖尿病肾病的风险预测提供一定的参考和借鉴。

Objective To establish 3 ensemble learning models for predicting type 2 diabetic nephropathy.Methods The type 2 diabetic nephropathy data were cleaned with several preprocessing methods based on the National Data Center of Clinical Medicine and their characteristics were analyzed by logistic regression analysis and mutual information score analysis respectively.The 3 ensemble learning models for predicting type 2 diabetic nephropathy,established by boosted decision tree,decision forest and decision jungle respectively,were trained for predicting the risk of type 2 diabetic nephropathy.Their efficiencies in predicting the risk of type 2 diabetic nephropathy were assessed and compared according to the accuracy,precision,recall,FI score and AUC respectively.Results The predictive value of the 3 ensemble learning-based prediction model of type 2 diabetic nephropathy was rather high.The prediction performance of the ensemble learning model established by two-class boosted decision tree was better than that of the other two models established by decision forest and decision jungle.Conclusion Ensemble learning-based diabetic nephropathy prediction model can provide certain references and experiences for predicting the risk of type 2 diabetic nephropathy.