Metrics

Sensitivity / Recall / True Positive Rate

Measure of the probability that the estimate is 1 given all samples whose true class label is 1, i.e., how many of the +ve samples were identified as being +ve (predicted & true +ve / all true +ve).

Specificity / True Negative Rate

Measure of the probability that the estimate is 0 given all samples whose true class label is 0, i.e., how many of the -ve samples were identified as being -ve (predicted & true -ve / all true -ve)

False Positive Rate

= 1 - Specificity = (predicted & false -ve / all true -ve)

Precision

Measure of the probability that a sample is true positive given that the classifier has said it is positive, how many sample predicted by the classified as positive as actually positive

ROC curve

• X-axis = False Positive Rate (= 1 - Specificity)
• Y-axis = True Positive Rate

Precision-Recall curve

• X-axis = Precision
• Y-axis = Recall

Use when +ve samples are very small compared to negative samples (highly imbalanced classes in samples)

References

• Quora: Difference between ROC curve and precision-recall curve
• Wikipedia: ROC curves