Predicting where crimes will occur

Welcome back. So we’ve talked about examples of how big data can be used to get quicker, more accurate measurements of how people are behaving in the world right now, and even potentially predict what they’re going to do in the future. We talked about examples in economics. But there are clearly a number of areas where it would be useful for us to have a better sense of what the future might hold. So say there had been a zombie apocalypse or some other kind of large disaster, which, amongst other people, had taken out the local chief of your police station. Let’s imagine somebody comes up to you and says, “well, I’ve heard you’ve been studying quite a lot recently.

You’ve been doing a lot of online courses. Would you be able to step in and help us and act as the local chief of the police station?” So now you have to decide where to put your police resources. So what are you going to do? A first thing you might think of is areas where you know a lot of crime has occurred in the past. Areas where people know it’s particularly dangerous there. This would be a reasonable first step. Obviously, looking at what’s happened in the past is often a good way of trying to get a better idea of what might happen in the future.

Now traditionally, police create maps, which they call hotspot maps where they plot where crimes have occurred in the past. And they colour in the areas where there’s been a particularly large number of crimes. However, some researchers at UCL, led by Kate Bowers and Shane Johnson, realised that if you have data, not only on where the crimes have occurred, but also when the crimes have occurred, you can possibly make better predictions. Specifically, what they found was that if a house had been burgled, then the chance that another house in that street would be burgled over the next two weeks increased significantly.

So if you know that a crime has occurred somewhere recently, then this is reason to believe there might be another crime there in the near future. So there’s a number of reasons that researchers think we might see these repeating patterns in crimes. One of the leading explanations is that burglars realise that there are items of value in a house, and they want to go back and get further items of value from that house, or from neighbouring houses. Or otherwise that burglars want to exploit the knowledge that they’ve accumulated about the layout of that house and potentially therefore also the neighbouring houses.

Now, some scientists in America have recently built on this idea. Have a think. Can you think of events in the natural world which are similar to the sorts of events we’re talking here, in terms of their repeating patterns? So, for example, we see one bad event and this bad event is followed by other, similar bad events. Some researchers who knew a lot about natural disasters realised that equations we use to better anticipate where earthquakes might occur, might also be of use in anticipating where crimes might occur. So, for example, if I asked you, where is it more likely that an earthquake might occur in the near future? San Francisco or London?

Sadly, that’s an easy question for us all to answer. We all know that San Francisco is in more danger of these sorts of problems than London is. But at the same time, we know that if an earthquake has occurred, it tends to be followed by aftershocks. And these researchers realised that the maths that we use to model these aftershocks and earthquakes and better anticipate where they might occur and when, can also be used to better anticipate where and when crimes might occur. And these maths are starting to be used in predictive policing software around the world.

In the following sessions, we’re going to have a look at how big data can be used to understand other kinds of crimes, and other kinds of conflict.