MachineLearningStatistics
Given a statistical model and a decision rule (parameterized by ) for two disjoint subsets , the ROC curve for a fixed parameter , is the graph for , of the test’s size against its power. If the ROC curve for one statistic lies above the another, it implies higher power for the same size of the test. This implies an decrease in Type II error (false negative rate) and an increase in recall for the same Type I error (false positive rate).