f1-score

Why 99% accuracy can be useless: how the confusion matrix, precision, recall, F1, and PR curves behave on imbalanced data, with a fully worked example.