Math algorithms are twice as successful as human analysts in predicting areas where crime will occur, according to a study released Wednesday by UCLA researchers.
In the Los Angeles part of the 21-month study that also examined a city in England, UCLA researchers pitted human crime analysts against mathematical predictive models to determine which is more accurate at predicting areas where burglaries, theft from cars and car thefts will occur in defined 12-hour periods.
“Not only did the model predict twice as much crime as trained crime analysts predicted, but it also prevented twice as much crime,” said UCLA Professor of Anthropology Jeffrey Brantingham, a senior author of the study.
Brantingham, who also co-founded of PredPol, a company that markets predictive policing software to cities, authored the study with UCLA mathematics professor Andrea Bertozzi and other scholars from UCLA, Georgia Institute of Technology and UC Irvine.
As part of the study, crime analysts at Los Angeles Police Department’s Southwest Division were asked to identify precise, half-block-size areas where crime will happen over a specific 12-hour span. The human analysts were correct 2.1 percent of the time, performing behind the mathematical models, which were accurate 4.7 percent of the time in identifying areas where crime will occur.
Researchers noted that while the success rates may seem small, the predictions were focused on “miniscule” target locations representing less than 1 percent of Los Angeles’ overall land area.
They also pointed to another phase of the study in which researchers tested the effectiveness of the predictive crime data in actually preventing crimes. Officers at three LAPD divisions — North Hollywood, Southwest and Foothill — were assigned to patrol 20 specific locations identified by either human analysts or the mathematical models. The officers, who were not told how the predictions were made, were instructed to respond normally to crimes.
The predictions made by the mathematical models resulted in 4.3 fewer crimes per week in the target area, or a 7.4 percent drop compared to the amount of crime that would normally be anticipated, while human predictions only resulted in 2.1 fewer crimes per week, according to the researchers.
Researchers also did a similar study in Kent, England, focusing on violent crimes such as assault. They also found that the mathematical algorithms were more successful than human analysts.
According to the researchers, the mathematical algorithms could save Los Angeles $9 million per year, based on the costs to victims, courts and society. They also said the mathematical models are designed to evolve and adjust to what it learns to become more accurate as time goes by.
The LAPD now uses mathematical models to predict crime in 14 of its 21 divisions, up from just three divisions using the models in 2013, according to UCLA.
The mathematical models used in the study were based on six years of research into 10 years of crime data.
The study was funded by grants from the National Science Foundation, the Air Force Office of Scientific Research, Office of Naval Research and the Army Research Office.
— City News Service