Boosting算法结合SMOTE技术在青年男男性行为者HIV感染预测中的应用

目的评价Boosting算法结合SMOTE技术预测青年男男性行为者(YMSM)HIV感染状况的性能。方法通过网络和现场抽取2018-2019年天津市YMSM 1179名,分别用XGBoost、LightGBM、CatBoost和logistic结合SMOTE技术建立预测模型,通过AUC、F1、Accuracy、Brier score等指标评价其分类性能。结果应用SMOTE合成数据后,logistic、CatBoost、LightGBM和XGBoost的AUC分别提升了23.4%、24.0%、25.4%和26.8%,Boosting算法的分类性能优于logistic模型。结论Boosting算法结合SMOTE技术为类不平衡数据的分类预测提供了新思路。

Application of Boosting Algorithms with SMOTE in Predicting HIV Infection among Young Men Who Have Sex with Men

Objective To evaluat the performance of Boosting algorithms combined with SMOTE in predicting HIV infection among young men who have sex with men(YMSM).Methods 1179 YMSM in Tianjin were selected through internet or site from 2018 to 2019.XGBoost,LightGBM,CatBoost and logistic combined with SMOTE were used respectively to build prediction models.The classification performance of models was evaluated by AUC,F1,Accuracy,Brier score and more.Results The AUC of logistic,CatBoost,LightGBM and XGBoost increased by 23.4%,24.0%,25.4%and 26.8%respectively after using SMOTE-based synthetic data and the Boosting algorithms outperformed logistic model.Conclusion Boosting algorithms combined with SMOTE provides a new way for prediction and classification of imbalanced data.