An Easy Way to Remember the Differences Between Precision & Recall
If you’re anything like me, at some point you have struggled to recall the differences between “Precision” and “Recall”. These are, of course, evaluation metrics used for classification tasks within machine learning. The two remaining major metrics are Accuracy and F1-Score, but this short post isn’t about them.
So how can we quickly and accurately remember what either of the aforementioned metrics means, say, in an interview setting? It’s simple. You only need to remember one sentence and the rest will follow logically.
That one sentence is: “Precision is linked to False Positives”. Once you memorise this, you can easily remember the equation for Precision, which is:
Mathematically, we know that if either of the terms in the denominator decreases, then the overall value increases. It follows that if we want to increase Precision, we must decrease the number of False Positives.
By making this association, we can easily remember the following too: Precision answers the question “of the number of instances which my model predicts to belong to a class X, what percentage of those instances actually belong to class X?”
Put another way, we can say Precision is given by:
“# positives correct/# labelled positive”
What about Recall? Easy. Since you remember that Precision is linked to False Positives, then it follows that the other measure, False Negatives, must correspond to Recall.
More specifically, if “FP” is in the denominator term of the equation for Precision, then “FN” must be in the denominator term of the equation for Recall.
If we want to increase Recall, it logically follows that we will want to minimise False Negatives.
And finally, if Precision is given by “# positives correct/# labelled positive” then recall is given by “# positives correct/# actually positive”
So, going back to the beginning, you only really need to remember the association between “False Positives” and Precision. Everything else follows through simple reasoning.
