Confusion Matrix Based AUC
AUC has now been taken as an important performance measure. For evaluating classifiers by AUC in the multi-class case, two extended versions of the measure can be exploited, in both of which, the AUC for each converted two-class sub-problem is computed in the same way as that of computing AUC in the twoclass case. Due to the general trait of multi-class classification, such AUC may mislead practitioners in ranking classifiers. In this paper, a new extension of AUC, called cmAUC, is introduced for multi-class problems. Different from the two extended versions of AUC, cmAUC is computed based on confusion matrices, emphasizing the probabilities of belonging to their true classes of samples. The analysis on synthetic data and the experimental results on some benchmark data sets demonstrate that cmAUC is relatively competitive for classifier evaluation.
classifier evaluation performance measure AUC cmAUC
Shuqin Wang Jinmao Wei
College of Computer and Information Engineering Tianjin Normal University Tianjin,China College of Information Technical Science Nankai University Tianjin,China
国际会议
太原
英文
145-148
2011-02-26（万方平台首次上网日期，不代表论文的发表时间）