🎹 Classification Metrics

Accuracy §

Accuracy is the main symmetric classification measure.

Precision and Recall §

Asymmetric measures include precision, recall (sensitivity or true positive rate), and specificity (inverse of false positive rate).

F1 Score §

F1 score combines precision and recall.

ROC Curve §

The ROC curve, pictured below, sorts predictions in descending order of confidence and measures sensitivity, the true positive rate, over a confidence threshold. As threshold increases, we predict more “yes,” so sensitivity increases.

Info

The threshold can also be interpreted as , or the false positive rate.

AUC §

The stronger the curve, the better the performance. Thus, area under curve (AUC) is another common metric for performance, varying between (random) and (perfect).

Confusion Matrix §

Finally, a confusion matrix shows counts of actual vs predicted class values.