PredPol – Predicting crime through data mining

Not too long ago in LA, crime was going up while the number of officers was going down. The LAPD had to try something different if they wanted to make a dent in this, so they looked to an anthropologist and mathematicians from UCLA, Santa Clara University, and UC Irvine.

“PredPol,” mines vast amounts of crime data and predicts where crimes will occur. Unlike the “hot spot” system, which identifies crime-heavy areas, PredPol is updated in real time and gives predictions for the next 12 hours. Cops in LA would go to these “boxes,” sometimes as small as 500 feet square, just to make their presence known and look out for criminal activity.

According to PredPol’s Proven Results page, the system was twice as effective as trained crime analysts. In the areas in which PredPol was tested, crime dropped by 13% while other areas showed a 0.4% increase.

PredPol works because, although an individual’s behaviour is very difficult to predict, once you put people in herds the trends and averages become very apparent. If you know the factors that contribute to a certain behaviour, you can work out a probability of that behaviour occurring. The more factors you know and the more accurately you know them, the better your prediction will be.

PredPol is being rolled out further, including the UK.

It’d be interesting to see how far you can take this. If you imagine a day where PRISM style data mining is legal and totally accepted, and governments can access all data, then combine that with “quantified self” monitoring (it won’t be long before neuro imaging become cheap and portable enough to be the latest personal informatics tool), you could pretty much predict anything, couldn’t you?


  • Dr Pullen says:

    Fascinating. In this age of big data raw computing power combined with great programming and imagination progress in areas we just need to open up to think about. Also runs risks like the concerns re governmental invasion of privacy. A two edges sword.

  • Interesting article. I assume that in order to feed data to the system researchers have to collect daily data from Tweeter and FB and run some analysis of sentiment or watch for specific alerts/triggers per geo location…?

Leave a Reply

Your email address will not be published. Required fields are marked *