Predictive analysis: It’s playing with probabilities

Predictive analysis: It’s playing with probabilities

SHARE

He describes predictive analysis as a place-based approach to crime analysis that utilises algorithm driven crime forecasts to inform decision making to prevent crime. In a tete-a-tete with GeoIntelligence, Captain Sean Malinowski of Los Angeles Police Department tells us why he thinks that technology holds the key for crime reduction in future…

Captain Sean Malinowski
Captain Sean Malinowski
Commanding Officer
Foothill Patrol Division
LAPD

<< He describes predictive analysis as a place-based approach to crime analysis that utilises algorithm driven crime forecasts to inform decision making to prevent crime. In a tete-a-tete with GeoIntelligence, Captain Sean Malinowski of Los Angeles Police Department tells us why he thinks that technology holds the key for crime reduction in future… >>

Q. You are the Commanding Officer of Real-time Analysis and Critical Response (RACR) Division. Can you tell us about your activities?
The RACR Division at Los Angeles Police Department (LAPD) is responsible for 24×7 situational awareness of what’s occurring in the city with regards to crime or emergency response. We also have another component at RACR which is dedicated to crime and investigative analysis and which maintains the official records or the number of crimes that have occurred. Moreover, we have a performance management command capability process which also provides support to detectives. We help detectives in case they are looking for an individual or want to identify a pattern, or require the assistance of the technology.

Q. What kind of technology are you using for carrying out your activities? How is it helping in controlling the crime?
We have an Esri-based platform for crime analysis mapping system. People can view the information about their area at www.crimemapping.com. For our internal use, we have a crime view server. For most of the things, we use CAMs (crime analysis mapping system). We have also got a bunch of individual databases and software that we use to search for patterns through those databases.

Captain Sean Malinowski Over time, we have been able to reduce the time between analysis and real-time incidents. This has been possible because we have detectives and officers monitoring the place 24 hours a day. So whenever a pattern emerges, we are immediately able to identify it. For example, recently we had an arson suspect who set 52 fires in a city over a couple of days. We were able to catch that person because we were operating 24×7 and had our resources in place to respond as and when required.

Q. You have been actively advocating for predictive analysis in policing? What is it and how does it work?
The reason that I have been pushing for predictive policing for the last few years is because we had a dramatic crime reduction in the city of Los Angeles and in many US cities that I feel we are now running out of space. At some point, we have to find something new because we have drained just about everything out of the existing programmes. We have become very good at these programmes, which are comsat based, hotspot policing, etc., over the years, and the history of crime has gone down by a few points, but now it’s getting increasingly difficult to get incremental crime reduction. That’s the reason why we are going to take the data that we have and use that for forecasting. The data will help us identify specific geographic areas where crime is most likely to occur. Over time, we will get more specific in time and place and deploy our officers based on the risk that we identify and forecast.

Q. You launched a threemonth trial programme in November last year for determining the effectiveness of predictive analysis. Were you satisfied with the results?
We are optimistic about the results so far. In three months, there’s nothing that we can say is statistically significant. We have thus extended the trial period to six months. The programme is certainly going in the right direction. We are seeing a crime reduction of about 8 per cent which we can contribute to the programme because we are doing a controlled randomised study. Then we are also finding that the algorithm is somewhere between 8-16 per cent more accurate in predicting what crimes would occur than we would without it.

Q. Is it feasible to ‘solely rely on computers to fight crime,’ as you have been quoted?
It is feasible. We are doing it right now. What we are doing is telling the officers, based on the results obtained from computers, where they need to be and when they need to be there. We are not telling them what to do when they get there. You really cannot do that. Depending on a situation, an officer will always use his knowledge, skill and experience to fight crime. We just GEOINTELLIGENCE Mar – APR 2012 26 assign them a box (an area). They are supposed to use their own problem solving skills to determine what is it about an area that makes it a high probability crime occurring area.

There’s a human element too. I think, officers and people get nervous when you say that you are assigning all the responsibility to the computer. What you are doing is allowing the computer with its lack of bias and computing power to turn to nearly 10,000 records of information and weigh it appropriately to screen out any background noise and point it in the right direction as to where the real patterns are.


Q. How is geospatial technology helping in predictive analysis?
None of this would be possible without geospatial technology. The existing system that we have enables geocoding of locations, which is very important. The model accepts the longitude and the latitude – so it’s only as good as the data we have. We have a very robust system in CAMS. On top of that, we have an algorithm programme which we have created. It generates the actual mission boxes on a base layer map.

Q. How do you think technology can aid in homeland security or fighting terrorism?
I know there’s some interest in homeland security community in how we can use this to, for instance, forecast where IEDs might be placed or for similar other activities. When we talk to mathematicians, they say it’s possible to use the technology for these purposes. We just need to understand the problem properly. I have spent years with the folks at UCLA, giving them data and having discussions with detectives to really understand the problem. Only then, we developed and tweaked the software.

I think it is doable. We can solve these types of problems with the technology and be able to come up with a forecast. However, we need to first understand the phenomena.

Q. Terrorists have a different way of thinking and operating than a criminal. So how can these kinds of solutions actually help in dealing with terrorists who are trained to kill people?
It may not be this exact algorithm. In fact, it probably wouldn’t be. But the theory behind it is based on human nature, not necessarily, the burglar behaviour. By human nature, I mean people, regardless of whether they are burglar or terrorist, do have routine activities that they are involved in.

These kind of activities fall into patterns. They probably will have different patterns from burglars, cheaters, but if you get enough data, you could identify what those patterns are and then turn them into some kind of actionable intelligence that could help in forecasting. And you know, it is not 100 per cent exact, it’s a probability – that’s all it is, it’s playing the probabilities. It’s better than random and in this case, it’s better than what we were doing before. And that’s why, what you can offer is a little better than what you were doing before.

We used to have people with lot of skills and experience doing forecast and deploying officers based on their analysis and the risk that they perceived. If you get computers to do that for you, it would be a great achievement. And what we are finding is that computer is doing this job a little better, so we are happy with that.