文献类型：期刊文献

中文题名：基于随机森林和多标记学习算法的慢性胃炎实证特征选择和证候分类识别研究

英文题名：Study on Feature Selection and Syndrome Classification of Excess Syndrome in Chronic Gastritis Based on Random Forest Algorithm and Multi-label Learning

作者：徐玮斐[1];顾巍杰[1];刘国萍[1];刘晏[2];颜建军[3];钟涛[3]

机构：[1]上海中医药大学基础医学院,上海201203;[2]上海市中医医院脾胃病科,上海310000;[3]华东理工大学机械动力学院,上海200237

年份：2016

卷号：23

期号：8

起止页码：18

中文期刊名：中国中医药信息杂志

外文期刊名：Chinese Journal of Information on Traditional Chinese Medicine

收录：CSTPCD;;CSCD:【CSCD_E2015_2016】；

基金：国家自然科学基金(81270050;81173199;30901897)

语种：中文

中文关键词：随机森林算法;多标记学习算法;慢性胃炎;特征选择;证候

外文关键词：random forest algorithm; multi-label learning; chronic gastritis; feature selection; syndromes

摘要：目的对慢性胃炎实证证候的特征症状进行选择,并建立证候模型,为慢性胃炎证候量化诊断的建立提供方法学参考。方法运用慢性胃炎中医问诊规范化量表采集临床症状和体征,并运用机器学习领域新提出的随机森林和多标记学习算法对慢性胃炎的实证症状进行选择和模型构建。结果运用随机森林和信息增益算法,结合多标记学习算法对证候分别建模,随机森林算法挑选出15个特征症状,信息增益方法挑选出20个特征症状,二者的模型最高准确率分别为83%、82%。通过评价,随机森林算法选出的特征症状更加精简,提高了诊断模型的识别率。结论随机森林结合多标记学习算法可实现慢性胃炎实证证候特征症状的选择,同时还可解决几个证候相兼问题,弥补传统学习算法的不足。
Objective In order to select the symptoms of excess syndrome of chronic gastritis, establish theclassification models and offer methodological references for quantitative syndrome diagnosis of chronic gastritis.Methods Normalized Scale of TCM Inquiry for chronic gastritis was used to collect clinical symptoms and signs.Random forest algorithm and multi-label learning algorithm proposed in machine learning field were applied to selectsymptoms of excess syndrome of chronic gastritis and establish classification models. Results Random forestalgorithm and information gain method were used and combined with multi-label learning algorithm to establishclassification models for syndromes of chronic gastritis. 15 featured symptoms were selected by random forestalgorithm, while 20 featured symptoms were chose by information gain method. The highest accuracy rates forclassification models were 83% and 82%. The evaluation results showed, the featured symptoms chosen by randomforests algorithm were more simplified, and the recognition rate of diagnosis model was improved. Conclusion Thecombination of random forest algorithm and multi-label learning can realize the selection of featured symptoms ofchronic gastritis, solve the problem of several syndromes occurred simultaneously, and make up the shortcomings oftraditional techniques.