不均衡数据集下急性阑尾炎患者住院费用预测

A prediction method of hospitalization expenses for patients with appendicitis on imbalanced dataset

目的通过对不均衡数据集的急性阑尾炎患者住院费用预测,探索提高少数类高费用患者住院费用预测精度的方法,为患者治疗方案选择、DRGs付费等政策调整提供理论依据。方法首先利用单因素分析识别显著因素,之后利用欠抽样和SMOTE过抽样对数据集进行均衡化处理,最后分别对原数据集I及均衡化数据集II构建Logistic回归模型。结果原数据集I构建的模型针对少数类样本的预测精度为60. 5%,而利用均衡化数据集II构建的模型针对少数类样本的预测精度达到89. 3%。结论利用均衡化数据集II构建的模型预测精度远高于原数据集I所构建的模型,有效提高了对少数高费用患者住院费用的预测精度。

Objective To explore the method of improving the accuracy of hospitalization expenses prediction for a small number of patients with high healthcare costs by predicting the hospitalization expenses of patients with acute appendicitis on imbalanced dataset,so as to provide a theoretical basis for patients’ selection of therapies and policy adjustments such as diagnostic related groups( DRGs) payment. Methods Firstly,single-factor analysis was used to identify the significant factors,and then random under-sampling method and synthetic minority over-sampling technique( SMOTE) were adopted to carry out balanced processing of the data sets. Finally,the logistic regression models were established using the original dataset I and balanced dataset II respectively. Results The accuracy of the model for the original dataset I was only 60. 5% for the minority class samples. And the accuracy of the model for balanced dataset II was 89. 3% for the minority class samples. Conclusions The accuracy of the model for balanced dataset II was far higher than the accuracy of the model for the original dataset I,which was more effective in accurately forecasting the hospitalization expenses for patients with high healthcare costs.