Datamining can often come up with a decent profile, especially if there is a large amount of information that is provided to the engine.

Picking on everyone that's brown is not a great idea as it's too crude. Indeed there may be other determinants that are far more important and might come to light.

Criteria should not be provided, merely the ones to check should be checked. Then a random sample of everyone else, as you've only managed to get the most likely based on past evidence.