ROC Curve

• ROC = Receiver Operator Characteristic
• ROC curve is used to compare one classifier with multiple parameters
• Usually used to find the threshold point of the output
• X-axis -> True Positive Rate or Sensitivity
• Y-axis -> False Positive Rate or (1 - Specificity)

• The point on the top is best for classifying positive instances but might not be good for negative one
• The point on the top left is the optimal one (TPR - FPR Tradeoff)