ROC analysis for predictions made by probabilistic classifiers
Zengchang Qin · 2005
Receiver operating characteristics (ROC) analysis was originated from signal detection theory and has been introduced to machine learning community in recent years to evaluate the algorithm performance under imprecise environment. ROC graphs have become increasingly popular in machine learning, because they offer a more robust framework for evaluating classifier performance than the traditional accuracy measure. In this paper, we investigate the relation between a probabilistic classifier and its corresponding predictor in a view of ROC analysis. A method of generating ROC curves for prediction (or regression) problems is proposed and some properties of ROC curves for prediction are discussed with examples.