Predicting crime using big data

Edition 20 June 2017, by Sagar Harinarayan

Imagine: A police officer is on patrol  duty. She is driving in a shady neighbourhood, with small groups of people in corners. Suddenly, her radio device starts beeping and a message displays, “Burglary is imminent on 24th street”. She stops the van and checks her email. There was a detailed description of a person who had been convicted of burglary on multiple occasions, in nearby areas. She swerves her vehicle round the corner and goes towards 24th street, looking for a face and sure enough, it appears. The man gives the officer a curt glance and keeps walking as if nothing had happened. But the officer knew something was surely averted and thought to herself, “Somebody can sleep well tonight!”

What happened in a few seconds was this: the camera captured an image of a man and uploaded it to the database. Here, a face-matching algorithm concluded that this man was a potential criminal. It understood the likelihood of burglary that night and was able to pinpoint an address based on the pattern of burglaries around that neighbourhood in the last year. This resulted in a precautionary email being forwarded to the police officer on duty.

Big data has revolutionized the service sector. Based on previous experiences, patterns are established to predict customers’ behaviour, causing a boom in b u s i n e s s . And it is only logical to utilize such analytics for people’s safety. The Los Angeles Police Department was the first to adopt this technique. Back in 2013, Professor Joel Caplan of Rutgers School of criminal justice highlighted that the approach described above keeps in mind shortterm objectives. Police officers are able to avoid criminal activities in a particular area, only to allow it to occur elsewhere. Alternatively, the potential criminals return once the police officers leave. Therefore, a more sustainable way would be to perform risk terrain mapping. The crime history of a certain region is merged with local behaviours to define crime-prone areas, thus taking into account impact of the environment as well.

Today, the Dutch police use this predictive algorithm extensively. It is called Crime Anticipation System (CAS) and continuously provides updates based on the current local conditions. Consider bike thefts, for instance. The Amsterdam police was informed that the probability of their occurrence shot up after 10 pm in a particular neighbourhood. This resulted in resources being deployed accordingly. Such trials in Amsterdam were extremely encouraging, and if outskirts were included, statistics showed that over 30% of thefts were committed in the zones predicted by the algorithm. This led to the technique being tested and validated in rest of the Netherlands.

The Netherlands Scientific Council for Government Policy (WRR) operates independently and provides the government crucial advice of a range of matters. WRR has investigated predictive policing thoroughly and made important recommendations. Typically, each zone analysed is 125 by 125 metres in area and the predictions hold valid for a two week period. Water and fields are not included in these zones. The CAS focuses on high impact crimes like home burglaries, assaults and street robberies. Simultaneously, it is being extended to others such as pickpocketing and business burglaries.

However, it is still early to comment on the effectiveness of CAS. One of the reasons being that the spectrum of crimes covered is limited. Murders and rapes cannot be predicted by it, given their rare occurrences. Besides, the system relies on data from the past and critics believe that the software does not take into account several inherent uncertainties. Therefore at the end of the day, it may simply lead to excellent police management, rather than crime prevention. I believe patience is key, as we are on track to unlock the complete potential of CAS.