基于机器学习煤矿工人肺通气功能障碍风险预测研究

Risk prediction of human lung ventilation dysfunction in coal miners based on machine learning

目的研究煤矿工人肺通气功能障碍的影响因素,通过机器学习算法构建矿工肺通气功能障碍发生的风险预测模型,为尽早识别肺通气功能障碍的高危人群、保护矿工健康状况提供重要的科学依据。方法选取2021年4月20日―5月3日在陕北某煤矿参加职业健康体检的679名矿工作为研究对象。通过非条件多因素logistic回归分析模型分析结果确定变量,构建逻辑回归(logistic regression,LR)、随机森林(random forest,RF)、支持向量机(support vector machines,SVM)和极端梯度提升树(extreme gradient boosting,XGBoost)模型并根据4种模型的准确度、灵敏度、特异性、阳性预测值、阴性预测值、F1评分、受试者工作特征曲线(receiver operating characteristic,ROC)下面积评估模型的性能。结果LR、RF、SVM和XGBoost模型的准确率分别为69.61%、70.59%、72.06%和75.49%。灵敏度分别为61.22%、58.16%、60.20%和64.29%。特异性分别为77.36%、82.08%、83.02%和85.85%。阳性预测值分别为71.42%、75.00%、76.62%和80.77%。阴性预测值分别为68.33%、67.97%、69.29%和72.22%。F1分数为0.66、0.66、0.67和0.72。ROC曲线下面积分别为0.78、0.78、0.78和0.81。XGBoost模型的预测性能优于其他模型,预测精度较高。结论运用XGBoost模型预测煤矿工人的肺通气功能障碍风险,为煤矿工人的健康管理提供相应的理论依据。

Objective The objective of this study was to explore the factors influencing lung ventilation dysfunction in coal miners and establish a high-accuracy predictive model using machine learning algorithms.This would aid in early detection of high-risk individuals and ensure better health safety measures for miners.Methods A total of 679 miners from a northern Shaanxi coal mine who underwent occupational health examination between April 20 and May 3,2021,were enrolled in the study.Using unconditional multivariate logistic regression analysis and Spearman correlation test to ascertain variables,we built logistic regression(LR),random forest(RF),support vector machines(SVM),and extreme gradient boosting(XGBoost)models.The models′performance was evaluated on metrics such as accuracy,sensitivity,specificity,positive predictive value,negative predictive value,F1 score,and area under the receiver operating characteristic(ROC)curve.Results The accuracy rates of LR,RF,SVM and XGBoost models were 69.61%,70.59%,72.06%and 75.49%,respectively.The sensitivity was 61.22%,58.16%,60.20%and 64.29%,respectively.The specificities were 77.36%,82.08%,83.02%and 85.85%,respectively.The positive predictive values were 71.42%,75.00%,76.62%and 80.77%,respectively.The negative predictive values were 68.33%,67.97%,69.29%and 72.22%,respectively.F1 scores are 0.66,0.66,0.67 and 0.72.The areas under the ROC curve are 0.78,0.78,0.78 and 0.81,respectively.Among all models,the XGBoost model exhibited superior performance,and the prediction accuracy was high.Conclusions The XGBoost model proved to be an effective tool in predicting the risk of pulmonary ventilation dysfunction in coal miners.This model could form a corresponding theoretical basis for the health management of coal miners.