Precision Recall curve is same as ROC Curve, with one difference that precision and recall is plotted.

• Precision is plotted on the Y axis
• Recall is plotted on the X axis
• Unlike ROC Curve, the top right-most point is the most optimal one
  • But that depends on the task
  • If its a cancer classification where False positive is okay, but false negative is not, there we have to use the point which have highest recall
  • But if its a spam classification where false negative is okay but false positive is not, there we have to use the point which have highest precision

Compare to ROC Curve

• Precision Recall is used if the the dataset is highly imbalanced
• Also, if some task wants high precision or high recall instead of the balance of sensitivity and specificity, then PRC is used

References