Abstract: AUC(Area Under the ROC Curve)evaluation criterion is widely used to measure the performance of classification algorithms in binary datasets. In this papaer, fistly we introduce the SVM Multi-class classification method; then we systematically introduce the AUC method; finally, we use the experiments to compare the AUC of various SVM Multi-class classification method in Multi-class datasets. The results indicated that the AUC value is closely related with the kernel functions and multi-class transformation method.

Key words: SVM, AUC, Kernel Functions, Multi-class transformation method

摘要： AUC(ROC曲线下面积)评价标准已经广泛地用于度量机器学习中各种分类算法在两类数据集上的分类性能.本文首先介绍了SVM(支持向量机)多类分类方法,然后对AUC方法进行了系统地介绍,最后通过实验来比较各种SVM多类分类方法在多类别数据集上的AUC的值.实验结果表明,AUC值和核函数和多类转换方法的选取都有着密切的联系.

关键词: 支持向量机, AUC, 核函数, 多类转换方法