摘要
本研究提出基于妊娠早期体检、基因信息,结合集成学习的妊娠期糖尿病预测分类方法.设计了基于Stacking框架的改进模型ACS-Stacking.ACS-Stacking模型将基分类器输出的类别概率值作为基层输出结果,元层使用GBDT模型学习组合基层输出的类别概率结果,拓展了算法的层次结构.在基分类器层与元分类器层之间加入基分类器筛选层,通过CFS算法估计不同分类器集合中个体分类器准确性与多样性的权衡值,筛选出最佳基分类器集合,实现基分类器的自适应选择.研究结果表明,该模型F1值较单一模型提高约9%,较Stacking模型提高约7%,具有较好的预测准确性和稳定性.
A prediction classification method of gestational diabetes mellitus based on early pregnancy physical examination,genetic information,and integrated learning was proposed in this study.An improved adaptive classifier selection-stacking(ACS-Stacking)model based on the Stacking framework was designed.In this model,the category probability value output by the base classifier was used as the output result of the base level.The meta layer used the GBDT model combined with the category probability results of base level output,which expanded the hierarchical structure of the algorithm.The base classifier screening layer was added between the base classifier layer and the meta classifier layer.The accuracy and diversity of individual classifiers in different classifier sets were estimated using the correlation-based feature selection(CFS)algorithm,and the best base classifier set was selected to achieve adaptive selection of base classifiers.The results showed that the F1 value of this model was increased by about 9%comparing with the single model,and about 7%higher than that of the Stacking model,which had high prediction accuracy and good stability.
作者
冯鑫磊
俞凯
袁贞明
FENG Xinlei;YU Kai;YUAN Zhenming(School of Information Science and Technology,Hangzhou Normal University,Hangzhou 311121,China;Engineering Research Center of Mobile Health Management System,Ministry of Education,Hangzhou 311121,China)
出处
《杭州师范大学学报（自然科学版）》
CAS
2023年第2期126-134,共9页
Journal of Hangzhou Normal University(Natural Science Edition)
基金
杭州市科技发展计划项目(20190101A03).
关键词
妊娠期糖尿病
基于相关性的特征选择
基分类器筛选
元分类器
gestational diabetes mellitus
correlation-based feature selection
base classifier selection
meta classifier
