A churn model is evaluated on 1,000 held-out customers, of whom 200 actually churned. At the default 0.5 threshold it flags 150 customers; 120 of them really churn.
What are the model's precision and recall?
Why did you pick that answer? Two or three sentences. The act of articulating it is what builds the judgment — not the click that follows.
Precision = true positives / flagged = 120 / 150 = 80%: when the model says "churner," it is right 4 times in 5. Recall = true positives / actual churners = 120 / 200 = 60%: the model catches 3 of every 5 churners and misses 80. Accuracy is (120 + 770) / 1,000 = 89% — which hides that 40% of the churners walked out unflagged.
Recall divides by actual positives (200), precision by predicted positives (150). 120/200 is recall, not precision.
120 / 1,000 = 12% divides by everyone. Recall only asks about the 200 who actually churned.
92% is close to the accuracy figure, not precision. Accuracy counts the 770 loyal customers correctly left alone.