Research on Measure Criteria in Evaluating Classification Performance
Feng Yan Qin, Cheng Ze-kai · Computer Technology and Development · 2006
The performances of classification models are different in data mining.How to select a good classifier is based on the evaluation of classifiers performances.Researches the measure criteria of classifiers performances in order to estimate classifiers effectively.The problems about the traditional measure criteria of classifiers performances are analyzed.ROC and AUC are introduced emphatically,and their virtue and shortcoming are anatomized.From the comparison and analysis,it shows that ROC and AUC are so attractive that they will be applied extensively,in spite of their shortcomings.