融合softmax的偏最小二乘法及中药数据分析研究

Analysis of TCM data with PLS method fusing softmax

偏最小二乘判别分析(partial least squares discriminant analysis,PLS-DA)是一种线性分类方法,不能充分表达数据之间的非线性关系,难以适应非线性数据的分类识别。针对该问题,结合softmax回归能够表达非线性特征,提出融合softmax回归的偏最小二乘判别分析算法(PLS-S-DA)。为了验证PLS-S-DA对非线性数据的有效性,以准确率、运行时间、查准率、查全率和F1-score为评价指标,采用四组UCI数据集和中药寒热药性数据集测试PLS-S-DA的性能,并与PLS-DA等五种分类算法对比。结果表明,对具有非线性特征的数据,PLS-S-DA相比于其他算法有更高的准确率,并对寒药和热药有更强的识别能力。

PLS-DA is a linear classification method,which cannot fully express the nonlinear relationship between data,and is difficult to adapt to the classification and identification of nonlinear data. Aiming at this problem,with softmax regression can express the nonlinear relationship,the paper proposed a partial least squares discriminant analysis algorithm fusing softmax( PLS-S-DA). In order to verify the validity of PLS-S-DA for nonlinear data,it used the accuracy,run time,precision,recall and F1-score as evaluation indicators. It used four UCI data sets and Chinese herbal medicine data set for testing. The performance of PLS-S-DA was compared with five classification algorithms such as PLS-DA. The results show that for data with nonlinear characteristics,PLS-S-DA has higher accuracy than other algorithms,and has stronger recognition ability for cold medicine and hot medicine.