3 ms·
One problem I see with this is that these statistics involve a lot of human bias. If you train your classifier on crime statistics, it's not based on a 'ground
by woodson 15y ago
One problem I see with this is that these statistics involve a lot of human bias. If you train your classifier on crime statistics, it's not based on a 'ground truth' (which is never attainable in a criminal case), but an expert's, i.e. judge's/jury's, decision.
- JoshTriplett 15y agoThat would indeed produce incorrect training for a classifier of "who should we pull out of line for a search". For that, you'd want to train exclusively on the observable characteristics of the people pulled out of line for a search versus whether you found something.