Crime-Predicting Algorithm Outperforms Humans

What’s the Latest Development?

Software developed for police forces, which predict where crime may occur in a given neighborhood, work better than human officers tasked with telling beat cops where to patrol. “The system produces, for each patrol shift, printed maps speckled with red boxes, 500 feet on each side, suggesting where property crimes—specifically, burglaries and car break-ins and thefts—are statistically more likely to happen. Patterns detected over a period of several years—as well as recent clusters—figure in the algorithm, and the boxes are recalibrated for each patrol shift based on the timeliest data.”

What’s the Big Idea?

The software, developed by a Santa Cruz start up and tested by the LA police force, was six times more accurate at predicting crime than choosing random patrol areas, and three times more accurate than human officers. “But whether the algorithm is right or wrong, it tends to reduce bureaucratic procedures and thus keep officers on the street, which by itself helps. … And if they do spook a would-be burglar into abandoning his plan, it means even more time on patrol, because the officer doesn’t have to leave his beat to process the suspect.”

Photo credit: Shutterstock.com