3 ms·
For imbalanced data, even accurate system will give tons of false positives. Unless you are willing to tolerate false negatives, having large false positive rat
by EvgeniyZh 3y ago
For imbalanced data, even accurate system will give tons of false positives. Unless you are willing to tolerate false negatives, having large false positive rate is almost inevitable. Rare disease testing is a popular example, but this one is a good instance to I think.
Let's say there are 10 million people linked to terrorism in the world and the system is correct 95% of times. Then out of 20000 people, 20 are positives, and 19 will be stopped; out of 19980 negatives, 999 will be stopped.
Compared to disease testing you may argue that FP is more harmful, but that's a different question. The point is that even very non-random system will be dominated by FP.
Edit: in particular, if they stop truly random people and 500 of them are linked to terrorism that implies that 20% of all people passing there are terrorists.
- burnished 3y agoProbably important to note that this leads to tests not being performed without good reason because they cause more harm than they do good due to the false positive rate.
- eastbound 3y ago> Unless you are willing to tolerate false negatives We tolerate false negative all the time: Terrorism is not eradicated at all in UK. Barring successful results, the police should stop harassing people.
- hardlianotion 3y agoI don't think we do tolerate false negatives, for this class of thing, all the time.
- sokoloff 3y agoTerrorists passed checkpoints today and we tolerated it. Terrorists passed checkpoints yesterday and we tolerated it. Terrorists passed checkpoints the day before that... The complete elimination of false negatives for rare occurrences is prohibitively expensive to the liberty of non-terrorists.
- EvgeniyZh 3y agoBeing unable to eradicate something is not the same as to tolerate it.