The problem with data mining is that it imposes the data miner's assumptions (let's call them prejudices -- pre-judgements) on the mound of raw data. Circumstantial facts can be used to build a damning pattern if you just spread the net widely enough, and given that it's estimated that 40% of all credit rating database entires contain bogus or factually inaccurate information, how do we know that we're not going to be dragged kicking and screaming into a SuperMax installation for debriefing on the basis of a typo in someone else's change of address letter sixteen years ago and a faulty parser grepping an ancient usenet posting spool?

Random example: back in 1989-90 I lived in Bradford, a city in the UK that is approximately 20% Muslim and has something of a reputation for radical mosques. I lived in a student house which was leased out repeatedly and owned by a local Pakistani landlord. And I've posted on usenet, including postings with keywords like "terrorism" and "al qaida" and "bomb" in them. If a Bad Type(TM) moved in across the street but their address was mis-keyed, or the date of their residence was mis-keyed, I think you can see where I might find myself coming up on someone's database search radar (for absolutely no valid reason).

Second random example: the problem of generalizing assumptions. Let us suppose my [fictional] younger sibling shares a flat with a gay friend who thinks they may be HIV positive, and I dig into my pocket to help out by buying books on living with HIV -- and even paying for an HIV test. The pattern of financial and other transactions on my record may make it look as if I have HIV (which is still something I'd be barred from entering the USA for), even though the truth was otherwise: after all, the medical records would be off-limits to most searches while credit records aren't.

This is just scratching the surface. If you want a classic example of the perils of blindly trusting automated procedures to catch terrorists, ask Teddy Kennedy about his time on the "No-Fly" list.