摘要
目的 采用随机森林的过滤式特征选择方法,筛选出最优特征子集后构建梯度提升决策树(Gradient Boosting Decision Tree,GBDT)模型,对平和体质与偏颇体质进行分类研究。方法 选取2756例受试者为研究对象,使用横断面调查法进行问卷调查。采集受试者十二经脉上24个原穴的穴位信号与身高、体重、年龄、性别等信息构建数据集。对数据集预处理后使用随机森林特征选择方法筛选最优特征子集;使用GBDT算法构建基于机器学习的平和-偏颇体质二分类研究;采用十折交叉验证计算准确度、精准度、召回率、F1得分,并综合评价模型性能。结果 筛选出22个特征构成最优特征子集,使用筛选后的特征子集构建的GBDT模型准确度、精准度、召回率、F1得分分别是92.86%、93.65%、93.08%、0.92。结论 随机森林的特征选择方法有助于筛选最优特征子集,GBDT可为中医体质分类研究提供帮助。
Objective To screen out the optimal feature subset and construct a gradient boosting decision tree(GBDT)model to classify peaceful constitution and biased constitution by using the filter-type feature selection method of random forest.Methods A total of 2756 subjects were selected as the research objects,and a cross-sectional survey was used to conduct a questionnaire survey.The signals of twenty-four original points on twelve meridians and basic information including height,weight,age and gender were collected and constructed as database.After the data set was preprocessed,a random forest feature selection method was used to filter the optimal subset of features,and then GBDT algorithm was used to construct a machine learning based pacific-biased body binary classification.And the calculation accuracy,precision,recall and F1 score were comprehensively evaluated by ten fold cross-checking,and the performance of the model was evaluated comprehensively.Results Twenty-two features were filtered to form the optimal feature subset,and the accuracy,precision,recall,and F1 scores of the GBDT model constructed using the filtered feature subset were 92.86%,93.65%,93.08%and 0.92,respectively.Conclusion The random forest feature selection method can help to filter the optimal feature subset,and the GBDT can provide help for traditional chinese medicine body classification studies.
作者
潘康宁
王洪杰
于霞
孙万晨
PAN Kangning;WANG Hongjie;YU Xia;SUN Wanchen(Department of Medical Equipment,Weihai Maternal and Child Health Hospital,Weihai Shandong 264200,China;Department of Ultrasound II,Weihai Maternal and Child Health Hospital,Weihai Shandong 264200,China;Medical Department,Weihai Chest Hospital,Weihai Shandong 264200,China)
出处
《中国医疗设备》
2024年第1期6-11,共6页
China Medical Devices
基金
山东省科技厅重点研发计划(2019GGX104078)。
关键词
机器学习
中医体质
特征选择
分类模型
machine learning
traditional Chinese medicine constitution
feature selection
classification model
