Metrics That Matter
Precision vs Recall
8 min+80 XPStep 1/10
Two questions, two metrics
Precision asks: of everything I flagged, how much was real? — TP / (TP + FP). For a spam filter: of the mail sent to the spam folder, how much was actually spam? High precision means few false accusations.
Recall asks: of all the real positives out there, how many did I catch? — TP / (TP + FN). For a cancer screen: of all actual tumors, how many were found? High recall means few misses.
Both matter. Almost never at the same time.
analogy
Precision is 'don't cry wolf.' Recall is 'don't miss the wolf.' Every classifier balances on that ridge.