Copyright © 2013 Location Based Policing. All Rights Reserved. Snowblind by Themes by bavotasan.com. Powered by WordPress.
I was pleasantly surprised last week when I saw a cable news network tease a new software program which would help cops predict crimes. I went and got a drink and set down in the recliner to see what the new software was. PredPol is a new crime prediction software that has helped reduce crime in Los Angeles and the city of Santa Cruz. The news blurb that I saw showed off some screenshots of the software and I must admit that I was intrigued.
One of the first things that jumped out to me about the software was that it utilized square boxes that highlighted a very small geographical area. Most programs out there nowadays utilize circles and/or ovals to highlight areas. I have long been a proponent of using irregularly shaped polygons to highlight crime areas. To me, crime does not occur in circles but in all sorts of shapes and sizes. If you confine yourself to just using circles then you have a lot of “dead-space” in which crime will not occur. Sometimes crime can’t occur in an area because there is a lake or some other geographical impediment. Unless your perp is out burglarizing houseboats on the open water, then why are we highlighting that sort of area to the troops?
The very small geographical area also surprised me. The predictive areas are 500′ by 500′! That is itty bitty in the world of predictive policing. The ability to be able to predict crime in such a small area is very advantageous to line level officers and commanders. I think back to a few years ago when I would routinely put out 1.5 mile buffers to patrol in. Being an ex-cop I know how difficult it is to patrol that big of an area adequately. If you can reduce the target area in size then the overall effort being put in to the patrol detail can be maximized. The key part to all this working is the accuracy of the predictions.
When I started to look into the algorithms being used for PredPol I became skeptical of the underlying premise. The algorithms are based off earthquake aftershock predictions. On face value this is a hard pill to swallow; what do aftershocks have to do with crime prediction? Fair question. Try as I may, I could not locate any of the research from the developers of the program as related to crime and aftershock correlation. However, I did dig up some of the research that the anthropologist had done prior (The key developers of the software are George Mohler – Mathematics Santa Clara University & Jeff Brantingham – Anthropology UCLA). Brantingham has done quite a few projects on the geographical correlation between geology features and crime. For example, geographical features should be taken into account when doing crime analytics because some features will affect crime. There will not be very many bank robberies on Mt. Everest. I speculate that Brantingham is suggesting that geological algorithms can be applied to crime because location does, indeed, affect crime. Hence the premise behind LBP.
The one thing I am concerned about in their model deals with the human behavior component. Earthquakes don’t have jobs and don’t have families to look after. Even though geography is a key piece of predictive algorithms, it is not the end-all/be-all factor. There is another great theory out there that takes into account geography and human behavior.
Crime is contagious. Did you know that? The near-repeat algorithms out there show us that crime is contagious and can act like a spreading disease under certain circumstances. Azavea Analytics is doing some ground-breaking work that ties in near-repeat functions with current incidents. I’ve been test driving some of the analytics over the past few weeks and I think it has potential. The whole idea behind basing predictive software off of contagion theories is that it takes into account human behavior. If I catch the flu then my actions dictate the locations of contamination. If I go to work my coworkers might get sick. If stay at home with my family then maybe they will get sick. Crime is the same way. If a burglar gets away with stealing in a certain neighborhood then he will come back. He is familiar with the neighborhood layout, streets, maybe police response times, opportune times for attack, etc… Crime is a disease that spreads in the form of near-repeat offenses. The contagion theory would tell us that if a perp (disease) comes close to being caught due to police intervention (police are like antibiotics in this analogy ) then this would modify how the perp commits future crimes. They probably won’t come back there because they almost get caught, thus your predictive hot spots would change. I’m not sure if the aftershock theory would take this into account as earthquakes are hard to intervene upon.
I applaud the work of groups like PredPol and Azavea. There are definitely good things on the horizon!
Recent Comments