摘要
为了利用ROC曲线下的面积(AUC),更好地评价多类SVM学习效果,提出了MOSMAUC(multi-objective optimizes multiclass SVM based on AUC)算法。该算法采用AUC作为评价标准,利用多目标优化算法作为SVM参数的优化方法,避免优化对象的AUC值过低问题,因为在多类分类学习中任何一个两类分类的AUC值太低,都会影响整体学习的效果。实验结果表明,提出的优化方法改进了算法的学习能力,取得了较好的学习效果。
In order to effectively apply AUC （area under the ROC curve） to do evaluation in multi-class SVM learning, an algorithm MOSMAUC （multi-objective optimizes multi-class SVM based on AUC） is proposed, where AUC is used as the evaluation criterion, multi-objective optimization is used to optimize learning parameters, since the low value of any AUC will decrease the learning performanee in the multi-class learning algorithm. Experimental results show the effectiveness of the proposed algorithm.
出处
《计算机工程与设计》
CSCD
北大核心
2009年第8期1960-1962,1973,共4页
Computer Engineering and Design
