基于AUC的支持向量机分类方法及应用研究

Research on classification method and application of support vector machine based on AUC

支持向量机(SVM)已经成为药物构效关系数据分析的一种常用统计方法,但其存在变量子集选择问题,且大量的冗余变量还可能影响SVM模型的预测精度,因此需要借助变量筛选来实现降维.本文提出了一种新的基于受试者工作特征曲线下面积(AUC)的支持向量机分类方法(AUC-SVM).首先,计算出变量的AUC值;其次,利用AUC值和前向选择算法选择最具信息量的变量子集,并剔除数据中无关和冗余的变量;最后,以AUC作为提取变量重要性的标准,通过真实的药物构效关系数据集来评估AUC-SVM性能,并与传统SVM方法相比较.实证结果表明,AUC-SVM算法能明显提高分类预测性能.

Support vector machine(SVM)has become a common statistical method for drug structure-activity relationship data analysis,but it has the problem of variable subset selection,and a large number of redundant variables may affect the prediction accuracy of SVM model,so it is necessary to reduce dimension with the help of variable screening.This paper proposes a new support vector machine classification method(AUC-SVM)based on the area under the subject working characteristic curve(AUC).Firstly,the AUC value of the variable is calculated;secondly,the variable subset with the most information is selected by using the AUC value and the forward selection algorithm,and the irrelevant and redundant variables in the data are eliminated;finally,taking AUC as the criterion for extracting the importance of variables,the AUC-SVM performance is evaluated through the real drug structure-activity relationship data set and compared with the traditional SVM method.The empirical results show that AUC-SVM algorithm can obviously improve the performance of classification prediction.