From medical diagnoses to smart assistants, artificial intelligence is helping us move from being a reactive world to a proactive one. That means machines that can anticipate events before they take place and make sure that the right precautions are taken. The world of policing is no different.
In an effort to better manage police resources and gain an upper hand in the war on crime, researchers from Georgia Tech and the U.K.’s University of Surrey have developed a new predictive policing algorithm — and it owes a debt of gratitude to technology that’s been previously used in weather forecasting and even the Apollo space missions.
“In our paper, we wanted to address two issues: ‘When is such an approach able to predict crime and when it won’t?’ and ‘How do we quantify uncertainty in when a crime will occur?’” Dr. David Lloyd from the U.K. university’s Department of Mathematics told Digital Trends.
The new algorithm built on previous work carried out by researchers from the University of California and police forces in both the U.S. and U.K. Their 2015 research showed how a predictive policing algorithm could accurately predict between 1.4 and 2.2 times more urban crime than specialist crime analysts. By making recommendations about where to patrol, the algorithm led to a 7.4 percent reduction in crime. However, while effective, this approach has also been criticized due to concerns about possible racial profiling and the underreporting of crime.
“In trying to address these issues, we needed to develop a new algorithm which achieved what the first algorithm does, but also captures uncertainty in the data and model,” Lloyd said. “For data that varies continuously — such as wind speed or spaceship position — there is a well-known algorithm, the Kalman filter, that does precisely this and efficiently fits a model to the data while capturing uncertainty. This algorithm was used on the navigation of the Apollo space missions and also for weather forecasting, where you have to process data in real time to make accurate forecasts.”
The new algorithm has so far been demonstrated on a data set of more than 1,000 violent gang crimes in Los Angeles carried out between 1999 and 2002. Early conclusions suggest that the upgraded predictive tool could prove superior for coping with the constantly fluctuating world of real-time crime prediction.
“There are several predictive policing software companies, such as PredPol.com and Hunchlab,” Lloyd said. “We would hope that our algorithms could be used in similar software.”
A paper describing the work was recently published in the journal Computational Statistics & Data Analysis.
