基于代价信息的二类分类器性能评估方法

Method for Appraising Performance of Two-Classifier Based on Cost Information

基于ROC曲线的AUC评估方法能有效评估二类分类器的性能,但是该方法只能评估分类器的总体性能,对代价信息不敏感。基于AUC方法提出用AUCCH方法评估二类分类器性能,该方法在具体代价信息下能分辨出最优分类器,在代价信息未知时能分辨出潜在最优分类器。在MBNC实验平台下编程实现,通过对AUC方法和AUCCH方法实验结果的比较,表明该方法具有有效性和健壮性。

The AUC method can effectively appraise performance of two - classifier, but it only can appraises overall performance of classi- tier and is insensitive to cost information. In this paper, a new method of appraising performance of two- classifier which is based on AUC is referred. The new method is called AUCCH. With cost information, the method can identify the most optimal classifier, With no cost information, it can identify the potential optimal classifier. Making experiment in MBNC experiment platform, through comparing the results between the AUC method and the AUCCH method, the results show that the new method is effective and robust.