基于机器学习的中医证素辨识算法模型集合研究

项目名称： 基于机器学习的中医证素辨识算法模型集合研究

项目编号： No.81202646

项目类型： 青年科学基金项目

立项/批准年度： 2013

项目学科： 医学八处

项目作者： 杨雪梅

作者单位： 福建中医药大学

项目金额： 23万元

中文摘要： "证素辨证"是一种"根据证候，辨别证素，组成证名"的新的辨证方法，是中医辨证模型研究中值得借鉴的研究成果之一，但该模型的广泛应用常因证素辨识准确率较低而成为瓶颈。本课题基于前期积累的逾万条中医临床案例，借鉴机器学习系统模型，研制若干中医证素辨识算法及模型（如人工神经元网络、决策树、最小二乘线性拟合等），在实现"症-证素"诊断权值动态修正同时，进行多个模型证素辨识准确率的大样本临床实验验证和比较。基于此，进一步尝试采用装袋法和推进法来汇总多个模型的辨识结论以达到提高中医证素辨识准确率的目标。当应用中模型集合辨识准确率急剧降低时，则激活算法集合重运行，构建新模型，实现模型集合及知识库的动态修正。课题研究成果一方面是具有更高准确率的中医证素辨识算法模型集合，一方面是中医证素诊断知识库，两者将为中医临床智能化自动辨证功能的实现提供核心部件，亦为证素辨证模型的大规模临床应用提供基础研究支持。

中文关键词： 中医证素辨识；机器学习算法；模型；爬山算法；支持向量机

英文摘要： "Differentiation of syndrome elements" is a new method that differentiates syndrome elements and gives syndrome names according to syndrome manifestations. It is one of the worthiest research results to be used for TCM(Traditional Chinese Medicine)syndrome differentiation model, but the wide application of the model was often limited by its low accuracy.Owing to more than ten thousand TCM clinical cases from our group's previous accumulations, this topic drew on the machine learning system model and built up some algorithms and models of TCM syndrome elements differentiation (including artificial neural network, decision tree, and the least square linear fitting) to verify syndrome elements differentiation accuracies of multipe models based on a large sample of clinical data as well modified the diagnostic weight values of "symptom-syndrome elements" dynamically. Furthermore, the bagging and boosting techniques were used to gather the identification conclusion of different models in order to improve the accuracy of syndrome elements differentiation. It'll activate algorithm sets to rerun for building new models in order to achieve dynamic modifications of model sets and knowledge base when the model accuracy is sharply reduced.One of the results is to find out a set of some algorithms and models of TCM syndrome

英文关键词： Syndrome elements differentiation of TCM；Algorithm of machine learning；Model；Hill-climbing algorithm；Support Vector Machine