Model Evaluation
Precision and recall are not just maths terms. They are product choices.
Use precision when you want your positive predictions to be mostly correct. Use recall when you want to catch as many real positives as possible.
Precision asks: out of the items your model marked positive, how many were truly positive?
Recall asks: out of all truly positive items, how many did your model catch?
0 of 7 lessons done · proved by a submitted hackathon project
Say your model flags college event posts that need moderator review.
It flagged 20 posts. 12 really needed review. There were 30 review-worthy posts in total.
precision = 12 / 20 = 0.60
recall = 12 / 30 = 0.40Do each one yourself, then tap it to tick it off. The ticks are only a checklist for you: they are not marked or scored.
0 of 5 done
For a UPI fraud alert model, recall is often more important. Missing a real fraud case is worse than checking one extra transaction.
For an auto-shortlisting model for internship applications, precision may matter if you show only 20 candidates to a recruiter. Bad recommendations waste limited review time.
Do not pick the metric after seeing which number looks better. That is result shopping.
Pick precision or recall before you tune the model. Then report the other one too, so people can see the tradeoff.
Add one plain sentence to your demo. It shows you understand the risk, not just the formula.
You can use this template in your README or presentation.
We optimise for recall because missing a real [positive case] is worse than wrongly flagging a safe case. We still report precision to show the false alarm rate.Do each one yourself, then tap it to tick it off. The ticks are only a checklist for you: they are not marked or scored.
0 of 5 done
Answer the quick check to finish this lesson.