Big data and machine learning harnessed for crime prevention

By October 09, 2015 1 comment

Japanese multinational Hitachi has developed a system which uses machine learning to predict crime.

A number of science fiction writers have invented technologies able to predict the future in order to pre-empt crime. The best-known tool in this genre is without doubt the software used by the New York police Pre-crime Department featured in Steven Spielberg’s film Minority Report, which is based on a short story by Philip K. Dick. However, the Pre-crime technology, which actually depends on psychic sensitive informants who are able to read the future in their dreams, is not very realistic. The fictional science of psychohistory, invented by Isaac Asimov, is somewhat closer to what tech firms are now starting to develop. Psychohistory is about foreseeing future events, in particular social phenomena, based on statistical analysis of huge volumes of data, which appears to presage the potential that big data processing is opening up today. In fact Japanese multinational Hitachi has just unveiled a new technology, called Hitachi Visualization Predictive Crime Analytics (PCA), which the company claims can predict crime in a given city by crunching large amounts of data and applying machine learning to it. According to the website of US business magazine FastCompany, the system is due to be tested in half a dozen United States cities starting this month.

Predictions based on crunching raw data

The technology developed by Hitachi uses a variety of data sources likely to have an impact on crime, including weather conditions, proximity to schools and subway stations, 911 (emergency) calls, gunshot sensors, population movements, and, of course, previous crime statistics. The data is then processed by machine learning, using a statistical software known as ‘R’ in order to find relevant patterns that humans might miss. Although there have already been a number of initiatives aiming to predict crime, so far not many of them have used machine learning. The traditional tools generally focus on a limited set of data that human observers think are relevant to potential crime. However, Hitachi’s technology excludes all human intervention: the machine crunches enormous quantities of data and delivers a predictive analysis. The results come from the raw data, without any human interpretation or hunches, which is claimed to make them more reliable.  "A human just can't handle when you get to the tens or hundreds of variables that could impact crime," Darrin Lipscomb, co-founder of Avrio and Pantascene, the companies which originally developed the crime-monitoring technology that Hitachi later acquired, told FastCompany.

Social networks an important source of information

PCA displays the results as colour-coded maps of the city in question, indicating the intensity of various crime indicators in each geographical zone. The system can pinpoint a location, down to a 200-meter square, and assign it a relative crime threat level from 0 to 100%. Analysis of data from the social networks plays a particularly important role in the prediction of crime. Lipscomb explains that most gangs use their own jargon on the social networks to plan their activities and that the algorithm can detect an abnormal term being used intensively in a given geographical area, which might herald potential problems. Some of the cities taking part in the pilot programme intend to relay their predictions directly to the police, so that officers can be dispatched to potential risk areas. Other cities plan to reveal the algorithm’s predictions only on completion of the pilot test, so that they can be compared with actual crime statistics gathered by the police over the period.

Page top

1 Comment

200 mètres carrés ça me semble compliqué. D'une part parce que c'est un niveau de détail très élevé, d'autre part parce que ça fait un côté de 14,14 m ce qui même en unités anglo-saxonnes ne représente rien.
Par contre un carré de 200m de côté, donc de 40000m2, soit 4 hectares, semble beaucoup plus réaliste !

Submitted by GuillaumeS (not verified) - on October 05, 2015 at 09:50 am

Legal mentions © L’Atelier BNP Paribas