利用LIME对脓毒症预测模型进行可解释性分析

Interpretability analysis of sepsis prediction model using LIME

针对机器学习应用于脓毒症预测存在预测准确率低和可解释性不足的问题,提出了利用LIME对基于机器学习的脓毒症预测模型进行可解释性分析。模型由预测和解释两部分组成:预测部分使用XGBoost和线性回归(LR),首先通过XGBoost进行特征提取,再利用LR对提取到的特征进行分类;解释部分使用LIME模型提取出关键的预测指标对模型进行解释。实验结果表明,通过XGBoost+LR模型进行脓毒症预测的准确率为99%,受试者工作特征曲线下面积(AUROC)为0.984,优于单独使用XGBoost(准确率:95%,AUROC:0.953)和LR(准确率:53%,AUROC:0.556)或者LGBM(准确率:90%,AUROC:0.974),同时通过LIME能有效地提取出前10个最重要的指标,对脓毒症预测模型进行可解释性分析,提高了模型的可信度。

Aiming at the problems of low prediction accuracy and insufficient interpretability in the application of machine learning to sepsis prediction,LIME(Local Interpretable Model-Agnostic Explanation)was introduced to interpret the machine learning-based sepsis prediction model.The model consists of two parts:prediction and interpretation.The prediction part used XGBoost(Extreme Gradient Boosting)and Linear Regression(LR).First,XGBoost was used to extract features,and then LR was used to classify the extracted features.The explanation part used the LIME model to extract the key predictive indicators and explain the model.Experimental results show that the accuracy of sepsis prediction through the XGBoost+LR model is 99%,and the Area Under the Receiver Operating Characteristic curve(AUROC)is 0.984,better than using XGBoost alone(accuracy 95%,AUROC:0.953)and using LR alone(accuracy:53%,AUROC:0.556)or LGBM(Light Gradient Boosting Model)(accuracy:90%,AUROC:0.974);and at the same time the LIME model can effectively extract the top ten most important indicators so that the sepsis prediction model can be explained and the credibility of the model can be improved.