A Better Measure than Accuracy in Classification Learning System
Feng Qin, Bo Yang, Zekai Cheng · 2006
Predictive accuracy has been used widely as a main evaluation criterion for predictive performance of classification learning system. However, it has many shortcomings and disadvantages, for example, it ignores probability estimations that classifiers produce. AUC (the area under the receiver operating characteristic curve) as a new measure of classification learning system is referred and recommended, which makes up for the deficiencies of accuracy and makes use of probability estimations or scores that classifiers produce. It is attractive and will be applied extensively. From the comparison and analysis, it shows that AUC is not only a better measure than accuracy but also should replace it in classification learning system