Mobile Phone Data Can Help Predict Crime a Month in Advance

Researchers used mobile phone and demographic data to predict London crime hotspots a month in advance.

Sep 19 2014, 1:30pm

It sounds like sci-fi fantasy, but advancements in big data analytics are allowing us to predict future crime trends with increasingly disparate and accurate methods. In a new study, researchers claim that they managed to predict crime hotspots in London a month in advance using just mobile phone and demographic data.

Instead of using background knowledge of the area, or profiles of offenders, their research uses "aggregated human behavioral data captured from the mobile network infrastructure," along with information about the people in the area such as education, ethnicity, language, and employment. The team, which includes members from Italy, Spain and the United States, report that their findings are accurate almost 70 percent of the time.

The "hotspots" they looked at aren't just a broad term that could apply to neighborhoods, but more localised, sometimes even applying to individual streets. As the researchers note, "In what are generally seen as good parts of town there are often streets with strong crime concentrations, and in what are often defined as bad neighborhoods, many places are relatively free of crime."

"The approach could have clear practical implications by informing police departments on how and where to invest their efforts."

The researchers cite previous efforts where mobile phone data has been used to model the spread of malaria; monitor citizens urban interactions; and understand socio-economic factors of different areas.

The data they used came from a public competition, the Datathon for Social Good, run by Telefónica Digital, The Open Data Institute and MIT during the Campus Party Europe 2013 at the O2 Arena in London.

The mobile data itself showed how many people were in a particular area, an estimation of their gender, age, and what they were doing there (chilling at home, working, or visiting). The data was already anonymised, and related only to the London Metropolitan Area. The data came from Telefónica's 'Smart Steps' product, which claims to provide "insights based on the behavior of crowds."

They also used a load of open data, including reported criminal cases, house sales, transportation, weather, and homelessness rates. In all, 68 demographic metrics were used in conjunction with the mobile phone data.

A map of predicted crime rates (right) compared to a map of real crime rates (left), where green is low and red is high. Image: Bogomolov et al

After taking all of these, cross-referencing them and linking them together, the researchers tried to predict if each area in the Smartsteps data would be a crime hotspot or not in the following month. To see how accurate their results were, the researchers checked their predictions against information from the "criminal cases dataset," which includes the geolocation of all reported crimes in the UK by month. Their model was successful nearly 70 percent of the time.

The researchers say that this approach could help police forces decide what areas of their jurisdiction they should focus on. "The proposed approach could have clear practical implications by informing police departments and city governments on how and where to invest their efforts and on how to react to criminal events with quicker response times," they write. But it could also act as a tool for investigating the underlying causes responsible for a certain area having a high or low crime level, which would be useful for policymakers.

As with all research, however, there are some limitations. The study dealt only with three weeks of Smartstep data, and the crime data was aggregated on a monthly basis, so it can only give a monthly forecast.

But this experiment nevertheless demonstrates the benefits of pooling multiple sources of big data together, in order to develop a better understanding of what's going on at a street-by-street level.