Fighting Crime With Big Data

Fighting Crime With Big Data

This post is also available in: heעברית (Hebrew)

by Konstantin Bodragin

Computers predicting crime have become a staple of science fiction films, but it could soon be a reality. The way it works will be somewhat different than what we’ve seen on the big screen.

The problem is that predicting the behaviour of individuals is hard. The way we act can appear random to an outside observer, no matter our internal logic . This makes the task of calculating our next move incredibly difficult for computers. When the focus is shifted, however, to a larger scale, the task becomes much easier.

Computer models are great at analysing patterns of activity of groups. When the sample is large enough – many thousands of individuals – groups of people behave more like an ant colony or particles, with areas of concentrated activity forming and dissolving over time. At this scale predicting collective behaviour becomes much easier. This is where Big Data comes into play.

In 2011 police departments in Los Angeles and Manchester, UK ran pilot tests of revolutionary applications of Big Data programmes to determine whether Big Data could be useful for reducing crime rates in specific areas. Turns out, it can.

The trial in LA was deemed a success, with a reductions of 12% in property crime, 21% in violent crime, and 33% in burglaries in the areas the software focused on. In Manchester, a reduction of 26.6% in burglaries in targeted areas, compared to a 9.8% reduction across the Greater Manchester area was recorded.

The London Metropolitan Police is currently conducting the biggest pilot of its kind in the UK. They are assessing three types of predictive technology and are expected to publish their findings later this year.

“Research shows predictive analysis can identify hotspots more accurately, and separate studies show targeting police patrol and problem-solving in hotspots can reduce crime. Forces in the UK and US are testing the effect of combining prediction with action to remove the causes of crime,” says Rachel Tuffin, director of research at the UK College of Policing.

In the Los Angeles trials, the system was fed data on 13 million instances of crime over the past 80 years. This allowed for better understanding of the nature of crime and the behaviour of criminals. The models showed that whenever a crime has been committed, more crimes were likely to occur in the vicinity. The patterns of criminal activity that emerged were similar to patterns of aftershocks. The models were surprisingly adept at predicting delinquency by relying on data from previous crimes.

Predictive technology may never prevent an individual crime from occurring. Nonetheless, Big Data will soon reach a point where it can assist policing officers by directing them to areas where crime is likely to occur. It will augment and can eventually supplant officers’ intuitions, and create an environment of shared operational experience – experience that many officers accumulate over many decades of dedicated police work.