Mastering Machine Learning with scikit-learn - Second Edition

Binary classification performance metrics

A variety of metrics exist for evaluating the performance of binary classifiers against trusted labels. The most common metrics are accuracy, precision, recall, F1 measure, and ROC AUC score. All of these measures depend on the concept of true positives, true negatives, false positives, and false negatives. Positive and negative refer to the classes. True and false denote whether the predicted class is the same as the true class.

For our SMS spam classifier, a true positive prediction is when the classifier correctly predicts that a message is spam. A true negative prediction is when the classifier correctly predicts that a message is ham. A prediction that a ham message is spam is a false positive ...