Search
• Antoreep Jana

# Understanding Precision vs Recall - Most Misunderstood topic

We'll begin with the definitions. Then we'll proceed to understandings. Finally, conclude with their applications.

Precision -> True Positives / ( True Positives + False Positives )

Ratio of the True positives by the total sum of (True Positives & False Positives)

Recall -> True Positives / (True Positives + False Negatives)

Ratio of True Positives by the total sum of (True Positives & False Negatives).

Understanding

We'll begin with understanding False Positives & False Negatives False Positives -> Predictions which indicated the Class Label as +ve but in reality were -ve False Negatives -> Predictions which indicated the Class Label as -ve but in reality were +ve

This means precision is an indicator of the samples or percentage of samples which were correctly identified as +ve among the total +ve predictions.

Whereas recall means the samples or percentage of samples which were correctly identified as +ve among the total +ve samples present in dataset.

If you didn't get the point, re-iterate the bold part and the last two sentences.

Applications Precision -> Use it as a metric when minimizing the False Positives is the focus. In other words, you don't want incorrect labelling as -ve class when the sample belongs to +ve class. eg. Advertisements -> You'd be more concerned with minimizing the prediction of a sample as unwanted which might be something the client might want. Rather than minimizing the prediction of a sample as wanted when it is unwanted. In the world of advertisements, majority of the stuff is unwanted. Loss is incurred when something which might be wanted has been labelled as unwanted.

Recall -> Use it as a metric when minimizing the False Negatives is the focus. In other words, you don't want incorrect labelling as -ve class when the sample belongs to +ve class. eg. Medical Detections -> A radiologist is more concerned with minimizing the predicting of a sample as -ve when the sample of the patient was +ve. Rather than minimizing the predicting of a sample as +ve when the sample of the patient was -ve. If the latter is done, it would be rectified in the several of the other medical stages. However, if the prior is done, a disease might go untreated. That's why Recall is used as the perfect metric by the radiologists.