Los Angeles is a strange juxtaposition of wealthy gated communities and crime-ridden neighborhoods. L.A. has fewer property crimes (with the exception of motor vehicle thefts) than the U.S. average, but it has higher numbers of violent crimes and homicides.
In recent years, the Los Angeles Police Department has taken a new approach to crime fighting. They’ve partnered with the University of California and a company called PredPol to use big data to predict where crimes might occur next.
So far, the city has seen a 33-percent drop in burglaries, a 21-percent decrease in violent crime and a 12-percent reduction in property crime, and the LAPD credits its data-driven approach. Despite recent questions about the quality of the LAPD’s data, early successes paint a rosy picture for big data and crime prevention.
Attacking Crime the Big Data Way
For years, Los Angeles has used data to predict where earthquakes might strike next. When major earthquakes rattle the area, predictive data can pinpoint the possible locations of aftershocks. A few years ago, the LAPD decided to feed its crime data into the city’s earthquake algorithm. After analyzing 13 million crimes and tweaking the algorithm a bit, they’re now able to use their data to predict where crimes might happen on a given day.
Data-driven crime prevention is not quite like the movie “Minority Report,” which relied on a network of clairvoyant humans who could predict crimes before they happened. Instead of predicting crime based on detecting violent human impulses, the LAPD’s algorithm assumes that future crimes, like earthquake aftershocks, are at least partly rooted in current crimes. Big data can reveal interconnections between families that might result in more crimes, and patrol officers can use certain data, such as truancy statistics, to predict where they should patrol. This kind of data implementation can work in harmony with current theories on criminal behavior to help law enforcement reduce crime rates. Although patrol officers resisted the change at first, they warmed up to big data when they noticed a significant reduction in crime.
Using data to prevent crime from happening requires police departments to feed accurate data into their algorithms. According to a recent Los Angeles Times investigation, the LAPD might be underreporting violent crime incidents.
Between September 2012 and September 2013, the LAPD classified nearly 1,200 violent crimes, mostly aggravated assaults, as minor offenses. For example, one aggravated assault in which a man stabbed his wife with a screwdriver and then shoved her down a flight of stairs was labeled as a minor assault. Another incident coded as a minor assault involved two men who beat another man unconscious with a metal crowbar.
The L.A. Times investigation casts a shadow over some of the department’s big data success. Mislabeled data means that the department underreported aggravated assault by 14 percent and underreported violent crime in the city by 7 percent. A civilian watchdog has launched an investigation into the reporting errors, which some police officers say were caused by relentless pressure to lower crime statistics.
Despite the discrepancies, the LAPD shows no signs of backing off of its big data strategy. In fact, Mayor Eric Garcetti plans to expand the use of big data throughout other Los Angeles public agencies.
Other Predictive Software Programs for Police
Predictive software can accomplish other police department goals as well. In Tennessee, an algorithm called C.R.A.S.H. (Crash Reduction Analyzing Statistical History) works to predict where traffic accidents are most likely to happen. It analyzes factors like weather, home football schedules, special events and historic crash data to predict motor vehicle accidents around the state. According to Tennessee police, the algorithm is accurate 72 percent of the time.
PredPol isn’t the only company that wants to leverage big data for crime fighting. BioCatch, a tool developed by an Israeli company, focuses primarily on preventing crime related to banking and e-commerce. The tool won’t replace traditional authentication methods, like PINs and passwords, but it can authenticate visitors upon return to a site, and it can flag users that demonstrate behaviors that are consistent with common fraudster behaviors.
BioCatch collects data on over 400 physiological, cognitive and bio-behavioral markers to create unique user profiles for bank and e-commerce customers. If someone logs into a bank account and doesn’t act like the account holder, BioCatch can note the problem and place a hold on the account. So far, BioCatch has raised $10 million in Series-A venture capital funding.
Image credit: CC by Airwolfhound