IE 11 Not Supported

For optimal browsing, we recommend Chrome, Firefox or Safari browsers.

Richmond, Virginia, Police Department Helps Lower Crime Rates with Crime Prediction Software

Virginia uses predictive crime analysis, data mining, reporting and GIS.

The Department of Pre-Crime is an intriguing concept used in the 2002 science fiction film Minority Report, and it's inching toward reality for the Richmond (Va.) Police Department (RPD). Instead of the film's fictitious "precogs" who float in a pool of water while foreseeing crimes, the police department uses crime analysis and prediction.

The concept is based on the idea that criminal behavior often follows identifiable patterns that can be used to predict criminal acts. By collecting yesterday's crime statistics and external factors - weather, time, day, moon phase, etc. - officers can estimate when and where tomorrow's crimes will occur using business intelligence (BI) capabilities.

Beginning in 2006, a new system was launched in a phased implementation that provides predictive crime analysis, data mining, reporting and GIS capabilities to the RPD. Officers receive the most current information available, including predictions of crime hot spots they can access before a shift. Data from the records-management system is integrated and analyzed continuously.

The RPD's innovative enterprise platform produced dramatic results. By moving from a "reactive crisis management structure" to a "proactive problem deference model," the department lowered the city's ranking from fifth most dangerous U.S. city in 2004 to 15th most dangerous city in 2005, and a 21 percent reduction in major crimes from 2005 to 2006. The department also won Gartner's 2007 BI Excellence Award.

"We're replicating the intuitive nature of the seasoned veteran cops - the guys who have been on the force for 25 years and know certain sections of the city really well and operate almost out of complete intuition - who know more than a crime map might show them," said Stephen Hollifield, information services manager of the RPD. "Our application attempts to do the same thing, but provides that type of intuitive picture about areas and gives them to our 'green officers' who have been on the force for only two years and haven't developed that sense yet. This kind of speeds up the process and gives pictures based on all the crimes in the past, weather, times of day, day of week and moon phases."


BI Approach Works

The RPD began down the road of crime prediction in 2005, when it named Rodney Monroe the police chief of its 700 officers. The immediate objective was lowering the Richmond's crime rate for its 220,000 residents, so Monroe met with BI software vendor Information Builders and analytical software vendor SPSS Inc. to see how technology could help.

At the time, the police department was data-rich and information-poor: A wealth of historical data was gleaned from its mature 911 system, the computer-aided dispatch system and the records management system, which were all used to track crime and ensure quality of service. But like many organizations, the police lacked a BI solution that could use the data.

Data has become a valuable resource for many government organizations, but its effective utilization - through tools like BI - has lagged behind. That's why BI has consistently ranked as a top priority for CIOs the last few years, according to IT research firm Gartner.

BI offers organizations invaluable system analysis by collecting, analyzing and integrating data, while providing historical, current and predictive views of business operations. Integrated reporting and analysis lets managers determine better management practices, improve services, identify effective strategies, enhance security and increase efficiency, among other things.

The RPD's first task was to identify data that would be used to create predictive crime reports - factors that wouldn't change drastically over short periods of time. This created a model that automatically improves itself and avoids the manual refreshing of variables. The chosen data included: time, day, holidays, weather, moon phases, city events, paydays and crime records. All the analyzed data was at least five years old to ensure the integrity of the predictions.


While the model places higher emphasis on the more recent data, the five-year window lessens our odds of predicting based on anomalies in any one data source," Hollifield explained.

Since there weren't packaged solutions that provided what the RPD required, the department developed a custom solution with several different technologies. The police data was integrated with Information Builders' WebFocus software, which included BI capabilities for analytics and reporting. The platform uses WebFocus as the primary user interface that displays criminal activity every four hours, which enables each new shift of officers to make adjustments for how they patrol. This information is available to officers at police stations and in squad cars; real-time alerts are sent by e-mail and text messages when extra officers are needed for deployment.

GIS mapping and dynamic geographical displays from ESRI and aerial photography from Pictometry International provide detailed dynamic pictures of the reported incident's location and the surrounding neighborhoods. The department used SPSS's Clementine and Predictive Enterprise Services product for data mining capabilities that examine how current crime reports relate to data on past, present and projected actions.

The system was integrated with, which feeds it contextual information about local activities, such as sporting events and the city's weather data collection system.


Crime Prevention Sophistication

The GIS capabilities let officers view specific types of crimes for a particular area and perform crime mapping and analysis functions. Officers can view maps of crime hot spots by location or crime type, such as car theft; they can also see specific incidents within a ZIP code, neighborhood, city district or other user-defined area. Data for weather, events, time, case history, associated suspects and aerial photos can also be integrated. The result is a sophisticated data model of criminal activity with a user-defined set of elements that predict criminal behavior.

From 2006 to 2007, the platform saw reduced incident rates of murder (32 percent lower), rape (20 percent), robbery (3 percent), aggravated assault (18 percent), burglary (18 percent) and auto theft (13 percent). A New Year's initiative used the platform to deploy officers in targeted areas, which resulted in a 49 percent reduction in random gunfire and a 246 percent increase in weapons seized - and only one-third as many officers were deployed compared to previous years, resulting in a $15,000 savings of overtime pay.

Analyses of criminal behavior also determined what crimes should receive higher priority. For instance, certain property crimes were better indicators of likely sexual assaults than the presence of convicted sex offenders.

"Now instead of looking at their static data, [officers] pull BI reports every two weeks and see where crimes are taking place," said Jeff Vining, research vice president of Gartner, who specializes in homeland security and law enforcement. "They wanted to be more predictive and have the officers there before a crime is committed; that's what makes this unique."

While metropolitan areas like Chicago, New York and Los Angeles use similar BI tools for crime analysis, the RPD is unique because it found its own solution with a limited budget, Vining said.

The RPD will soon roll out the next phase of the application by moving to a more granular approach that predicts the intensity of crime within four-hour windows via seven- and 30-day forecasts. The system will categorize crimes as nondomestic, robbery, burglary, auto theft, theft from auto and all other larceny. With different strategies and tactics to address each type of crime, the new model will provide more intelligent analysis for the RPD's precinct commanders, Hollifield said.

"This is going to give us a prediction of what to expect over the next seven days and what to expect broken out in four-hour windows," Hollifield said. "In that four-hour window, we can predict what particular crime type could occur and the probability of that crime happening."