Feature - Predicting burglary with the grid

Photo courtesy Andy Fox, stock.exchng

Superheroes like Batman are not the only ones who can make use of sophisticated technology to fight crime. Nick Malleson, a researcher at Leeds University, has designed an intricate computer model to forecast burglary rates, which relies on the UK’s National Grid Service (NGS) to provide the necessary computing power.

Predicting crime is a tricky business, because the likelihood of a burglary can depend upon numerous human and environmental factors, all of which affect one another. In an attempt to forecast general trends, Malleson’s model simplifies the complexities surrounding crime prediction by using an “agent-based” model — one in which largely autonomous individuals, or “agents,” make decisions and perform actions which are influenced by the multiple individual factors within their environment.

In his model, potential burglars make decisions about whether to work, sleep, socialize, take drugs, or resort to burglary. (The legal definition varies, but burglary is generally defined as entering a building with the intention of committing a crime, such as wrongfully taking property.)

Adding or subtracting different factors affects how burglars respond to their environment. For example, things like established, close-knit communities and security systems on houses should, according to common belief, discourage burglars from going ahead with the crime. Meanwhile, some have the perception that the chance of burglary increases when easily accessible, widespread public transportation is readily available. Malleson’s model allows one to test these hypotheses.

Furthermore, his model can fine-tune factors, testing how much of an impact a change in these environmental differences makes to a would-be burglar. For example, a theoretical new condition such as a new bus route or housing complex can be introduced, and the model will calculate what consequence this would have for the overall burglary rate throughout the city of Leeds.

The model cannot be used to forecast other types of crime such as murder or rape, but it is adaptable to other cities and countries, provided that the required local data is available. Malleson has already successfully modified it to make predictions for Vancouver when visiting expert criminologists at the Institute for Canadian Urban Research Studies. Here, the model revealed that large urban transportation centers showed high crime rates regardless of whether the agents used public transportation or not, indicating that the physical layout alone of a city might affect the crime rate — adding geography as yet another, unexpected, factor to the prediction of crime.

Two not-so-model citizens: forecasting where burglars stalk the streets (click on image for a closer look.) GeoTime(r) software used courtesy of Oculus Info Inc. All GeoTime rights reserved

The complexity of crime

Due to the complexity of the system, each model takes 20 hours to run on an ordinary desktop PC. Worse, due to its probabilistic nature, about 100 models need to be calculated before reliable results can be produced.

The NGS offered a critical solution to speed up the process. Because the model is written in Java, it can be run on NGS directly. All Malleson had to do was adapt the program so that it could exploit free nodes on the NGS simultaneously, allowing him to run several models at once.

“Although individual NGS nodes aren’t much more powerful than a desktop PC, I can utilize up to 128 nodes simultaneously, so it is possible to get 128 results in the time it would normally take to get one result on a single PC,” explained Malleson. “This is where the NGS was essential for the project to be feasible. Without access to NGS resources, the project would not have been able to continue.”

Safer Leeds, the local crime and disorder reduction partnership, is already interested in the project, and it has been working closely with Malleson to provide him with expert knowledge and data. In the future, the model could influence their real-life policies and ultimately be used to help predict and reduce burglary rates.

—Seth Bell, iSGTW. Want to comment? See our Nature Network forum.