There are two error types in classification problems: False positives (FP) and false negatives (FN). I find it more convenient to think in False Alarms (FP) and Missed Hits (FN).
Example: “Is a person infected?”
→ classify each person into infected/healthy
- Error 1 (FP), False Alarm: classify “infected” but is healthy
- Error 2 (FN), Missed Hit: classify “healthy” but is infected
Classification tests can be tuned to favor one error over the other:
- Favor false alarms: HIV test → You do not want to miss an infected person
- Favor missed hits: Airbag trigger → You do not want an exploding airbag in your face when driving normally
For example, the sensor in the car for triggering the airbag is designed to favor missed hits but almost never produce a false alarm. Thus, the airbag may only trigger 50% of the time when having a bad accident but should never trigger when bumping into a pot hole. The rationale is that an airbag that unexpectedly explodes into your face during normal driving leads to accidents in over 50% of the time.