Sarah Rich, Government Technology, August 19, 2011
The Santa Cruz, Calif., Police Department implemented a six-month predictive policing pilot project, which began July 1, to help officers predict certain types of crime in the city before it happens.
Through the predictive model, officers will patrol areas that weren’t previously receiving enough of a police presence with the goal of deterring crime.
The project uses an algorithm that is similar to what’s used for predicting earthquake aftershocks. “There’s a belief that certain crime types–in this case, burglaries and vehicle thefts–can be predicted in the same way,” said Zach Friend, the Santa Cruz Police Department’s press information officer and principal management analyst.
For the six-month pilot, the Police Department pulls crime data every day from its record management system that tracks crime that’s been reported in the city. The data is put into a spreadsheet and geo-coded and then run through Mohler’s Web-based computer algorithm.
The result is 10 maps outlining Santa Cruz’s crime hot spots, which are distributed to police officers, who then can patrol more efficiently based on that information.
In the nearly two months of use, the pilot has garnered positive results. Since the pilot’s deployment, the model has correctly predicted 40 percent of the crimes that it was aiming to predict, and the Santa Cruz Police Department has seen a reduction in the types of crime that it’s been addressing.
In addition, the Police Department saw a 27 percent decrease in the number of reported burglaries in July compared with July 2010. Friend said the department won’t know how successful the model is until it’s been running for at least three months.
Santa Cruz’s model also removes potential biases officers may have about a particular area they patrol, according to Friend. If an officer has patrolled a certain neighborhood for a few years and is aware of problematic homes with inhabitants that have a history of drug use or criminal activity, that officer may feel inclined to spend additional time going by those locations.
“The model normalizes the information. It doesn’t look at people, it simply looks at crime,” Friend said. “[The model] may reinforce that you should go back to the [problem] area, but maybe only twice that week as opposed to all four days that you work your shift.”