主成分-线性判别分析在中药药性识别中的应用

Discrimination of properties of Chinese Traditional Medicine with principal component analysis-linear discriminant analysis

目的探索利用主成分-线性判别分析基于中药的功效主治属性判别其药性的可行性。方法收集《中华本草》中收录的药性明确、功效主治属性特征详尽的植物药1 725种,首先运用主成分-线性判别建立模型,以此模型对药性进行判别分类,采用10次5折交叉验证评价模型稳定性,然后按照随机抽样的原则,从寒、热性两类药材中分别随机抽取80%(共1 380种)的药材作为训练集建立模型,其余20%(共345种)药材组成测试集,做预测。结果运用主成分-线性判别模型,全部1 725种中药的组内判别正确率为94.43%,交叉验证平均正确率为91.54%。训练组的组内回代判别正确率为94.78%,测试组的预测正确率为90.14%。结论基于主成分-线性判别对中药药性进行判别,不仅保证了线性判别的正常运行,而且判别准确率高,模型稳定性好,能够为临床用药提供依据。

Objective To identify the feasibility of principal component analysis-linear discriminant analysis（PCA-LDA） to discriminate properties of Traditional Chinese Medicines（TCM） based on their efficacy and indication characters.Methods Information on efficacies,indications,and properties of 1 725 kinds of TCM was collected from ＂Chinese Herbal Medicine＂.PCA-LDA was applied to construct a model and to discriminate properties of TCM based on efficacies and indications of 1 725 kinds of TCM.10 times 5-fold cross-validation was used to evaluate the stability of this model.The overall data was randomly divided into two subsets： 80% of the data（1 380 samples） from TCM of cold nature and hot nature were used as the training set to construct the model,and the remaining 20%（345 samples） were used as the testing set to evaluate the prediction accuracy.Results According to the PCA-LDA model,the consistent accuracy was 94.43% for 1 725 kinds of TCM,and the mean accuracy of 10 times 5-fold cross-validation was 91.54%.The consistent accuracy in the training set was 94.78% and the predication accuracy in the testing set was 90.14%.Conclusion PCA-LDA discriminating properties of TCM could insure linear discrimination and guide clinical prescription with high discriminant accuracy and stability.