Model evaluation for classifiers: precision, recall, ROC and imbalanced data
A fraud-detection model with 99.13% accuracy sounds impressive, until you learn that a program that prints "not fraud" for every transaction scores 99.12%. That is why model evaluation for classifiers...











