基于机器学习算法的医院职工满意度影响因素研究  

Research on influencing factors of hospital staff satisfaction based on machine learning algorithm

作　　者：胡鹏[1] 杨媛 黄艳[2] Hu Peng;Yang Yuan;Huang Yan(Hospital Office,Guangzhou First People′s Hospital,Guangzhou 510180,China;Department of Science,Education and Information,Guangzhou First People′s Hospital,Guangzhou 510180,China)

机构地区：[1]广州市第一人民医院办公室,广州510180 [2]广州市第一人民医院科教信息部,广州510180

出　　处：《中华医院管理杂志》2024年第11期862-869,共8页Chinese Journal of Hospital Administration

摘　　要：目的引入机器学习算法建立医院职工满意度影响因素研究模型,并借助沙普利可加性特征解释方法(SHAP)对模型进行解释,以期为提高医院职工满意度提供参考。方法2022年12月,采用自制问卷对广州市某三级甲等医院的职工进行满意度调查。对问卷数据进行描述性分析和单因素方差分析,使用随机森林筛选特征,在十折交叉验证的基础上采用自适应提升算法、极端随机树算法、梯度提升算法以及极端梯度提升(XGBoost)算法建立职工满意度预测模型,通过混淆矩阵和受试者工作特征曲线下面积(AUC)评估模型的正确率、精确率、召回率和F1值,利用网格搜索技术并以AUC作为评价指标对性能最优模型进行参数优化。运用SHAP方法对最优模型进行全局和局部解释。结果共收回有效问卷814份,职场关系和后勤管理两个维度的得分较高,评分均值分别为3.96分和3.91分,工作强度维度的得分最低,评分均值为2.92分。职工总体感受为满意者536人(65.85%),不满意者278人(34.15%)。不同年龄、岗位类别、职称和工作年限的职工的满意度总体感受差异具有统计学意义(P<0.05)。XGBoost算法模型在测试集上的AUC值、正确率、精确率、召回率和F1值分别为0.9435、0.8944、0.9320、0.9057和0.9187,综合性能优于其他算法构建的模型,且模型拟合效果好;参数优化后,AUC值、正确率、精确率、召回率和F1值分别为0.9608、0.9006、0.9327、0.9151和0.9238。SHAP方法分析结果显示,对职工满意度影响最为显著的因素按重要性程度排序依次为解决问题、公平回报、晋升对比、福利对比和院务公开。结论XGBoost算法结合SHAP方法解释框架可用于揭示和剖析影响医院职工满意度的关键因素,可为医院管理者有针对性地提高职工满意度提供更精准的决策支持。Objective To establish a research model of influencing factors of hospital staff satisfaction with machine learning algorithm,and to explain the model with the help of SHapley additive explanations(SHAP)method,in order to provide reference for improving hospital staff satisfaction.Methods In December 2022,a satisfaction survey was conducted among staff of a tertiary hospital in Guangzhou using a self-made questionnaire.The questionnaire data were analyzed by descriptive analysis and one-way analysis of variance,and random forest was used to screen features.On the basis of ten-fold cross validation,the staff satisfaction prediction model was established by adaptive boosting algorithm,extreme randomized tree algorithm,gradient boosting decision algorithm and eXtreme gradient boosting(XGBoost)algorithm.The confusion matrix and the area under the receiver operating characteristic curve(AUC)were used to evaluate the accuracy,precision,recall and F1 score of the model.The grid search technique and AUC were used as the evaluation index to optimize the parameters of the best performance model.The SHAP method was used for global and local interpretation of the model.Results 814 valid questionnaires were collected,the scores of workplace relations and logistics management were higher,with an average score of 3.96 and 3.91,respectively.The score of work intensity was the lowest,with an average score of 2.92.536 staff(65.85%)were satisfied with the overall feeling,278 staff(34.15%)were not satisfied.The difference in the overall feeling of satisfaction among staff of different ages,job categories,titles and years of experience was statistically significant(P<0.05).The AUC value,correct rate,precision rate,recall rate and F1 value of the XGBoost algorithm model on the test set were 0.9435,0.8944,0.9320,0.9057 and 0.9187,respectively,which were better than those constructed by other algorithms in terms of comprehensive performance,and the model was well-fitted;after parameter optimization,the AUC value,correct rate,precision r

关 键 词：人员管理 医院 机器学习 极端梯度提升算法 沙普利可加性特征解释方法 医院职工 满意度 

分 类 号：R197.323[医药卫生—卫生事业管理]