It's 3 a.m., and you and a few other officers are on surveillance. The city has been plagued by a rash of hot-prowl burglaries over the past few weeks. The suspect has not been seen, usually because the victims were asleep when the burglary occurred.
The Crime Analysis Unit predicted the perp might strike in this area between the hours of midnight and 4 a.m. They also suggested he might be on foot or bicycle. His modus operandi is to remove louvered window panels or enter via unlocked doors, so he travels light and carries few tools. He likes to strike first-floor structures.
At 3:25 a.m., the radio silence is broken. "I see movement at post No. 3," one of the officers says. "One L31, the suspect just entered via a louvered window."
At 3:30 a.m., one of the officers reports: "One L32, code four, suspect in custody."
Any law enforcement agency could be the problem solver in this scenario by using predictive analytics to solve crimes before they occur again. In this instance, the local police crime analysis unit was able to process police report data from past incidents to predict when and where the next crimes were likely to occur. The watch commander only had to strengthen police staffing in targeted areas, and wait.
Advanced predictive analytics could be the next generation problem-solving tool for the policing profession. Predictive analytics combines existing technologies like computers, crime analysis and well-developed police reporting techniques and adds a few newer technologies such as artificial intelligence, universally shared data and borrowed technology from the consumer industry to build a system that is capable of predicting crime before it happens. The result is a network that gathers disparate data and uses software that is engineered and tested in the private customer management industry to model patterns and trends from crime data.
But high-tech tools alone cannot solve crimes. Crime analysts within the agency must analyze the results and offer police management insight as to where police resources are likely needed the most. And managers must balance their daily needs for patrol services with the potential to catch criminals through the strategic placement of officers in the field.
Predictive crime analysis packs potential
Law enforcement agencies have contributed crime data to the FBI's Uniform Crime Reporting (UCR) system since the 1930s. Participation in this system is not mandatory. However, as of 1995 the statistics represented in the UCR program included 95 percent of the nation's total population. The ability to quantify crime incidents in this way is one of the reasons predictive analytics presents such an excellent option for problem solving.
Consider that: According to the FBI Crime Clock (2006), residential burglaries occur once every 14.4 seconds in the United States. Of the roughly 2,183,746 residential burglaries reported nationwide, little more than 275,000 were cleared by arrest. These numbers alone seem to present a huge opportunity for policing to get ahead of the curve, and actually make a dent in reducing or eliminating property crimes altogether. The FBI Crime Clock also reported crimes of passion such as homicides occurred much less frequently; at an average of one every 30.9 minutes. In either case, it is possible to predict certain drug-related or domestic violence-related homicides because of the meticulous counting law enforcement does, reports Colleen McCue, author of "Doing More with Less — Data Mining in Police Deployment Decisions."
Leveraging computer tech
Fueled by the confluence of evolving computer technology, improved sharing of data between agencies and adaptations from the private sector, it is possible to bring reliable predictability and advanced problem solving to local police agencies. It's a matter of simple averages that computer technology alone will solve some crimes, and so will improved data sharing between law enforcement agencies. What modern policing needs to understand is that advanced predictive analytics acts as a force multiplier when all of these technologies are focused on problem solving.
However, McCue reminds that the use of computers alone will not solve U.S. crime problems. Policing, she says, needs to look beyond its traditional framework, and consider technology sharing from sources outside police work.
Fortunately, the indexing of data among different data sets is straightforward in policing today. Unlike the filing cabinet systems of old, electronically accessing data is easy. Officers can retrieve data virtually, so they don't need to go to a file cabinet to get it.
And sharing of data among departments is simplified as well. Universal formatting is slowly becoming a non-issue, as the computer industry has adopted standards at many levels and within most disciplines. The Windows XP and Vista platforms are examples of how Microsoft has standardized many aspects of the PC market. Within the justice community, the Global Justice Extensible Markup Language Data Model (GJXDM) has been adopted to ensure agencies can share data among and between data systems.
Determining data access
The Customer Relationship Management (CRM) industry focuses on predictive behaviors in the consumer industry. This industry has pioneered work with companies like Amazon.com or the local grocery store. When you last logged onto Amazon.com and ordered an item, did you see something to the effect of "other customers who placed similar orders also ordered these items?" At your grocery checkout have you received coupons based on your purchases? The CRM industry is huge, and may be considered an experienced partner in harnessing social sciences to predict human behavior.
The processes involved in predicting your shopping behavior are the same ones used to determine criminal behavior, just using different data sets, reports McCue along with authors Emily Stone and Teresa Gooch in "Data Mining and Value-added Analysis."
However, keep in mind that private industry may be able to capture more personal data than a government body such as a police department. This is based on the perception of what police are likely to do with this personal information, so it is tightly regulated. The State of California establishes certain personal data as Criminal Offender Record Information (CORI), and strictly limits access to that data on a "need to know" and a "right to know" basis. The first authorizes access to certain records by statute. The second definition includes information that is required for the performance of official duties, according to the California Department of Justice.
Inputting and sharing data
Once authority to access data is established, then one must ask how much data exists and who is going to enter it.
Police and public safety computer data systems should be smart enough to import data already entered once into a computer to eliminate redundant entries. For instance, when a subject is booked then cited out for the same violation, all of the booking data should be "pre-filled" in the fields for a citation entry screen. Then the officer only has to make simple changes to complete the citation.
On a larger scale, when data collected at one agency is not shared with another, it is both inefficient as well as impractical. Inefficient because personnel costs are a large part of what drives data entry costs and they are limited, expensive and slow. Impractical because "The average company's data storage needs triple every 18 to 24 months, and the worldwide data storage capacity has grown from 283,000 terabytes in 2000 to nearly 5 million terabytes by 2005," according to the GE Global Research report "Holographic Data Storage." With data growth advancing so quickly, only computer systems will have the power to wade through massive amounts of information and help individuals distinguish relevant information from garbage, note McCue, Stone and Gooch.
Problem solving — the new state
As a group, law enforcement is good at counting its crime data, as evidenced in the annual reports submitted to the FBI by most law enforcement agencies. Departments have become more sophisticated in their approach to technical solutions, and are much better at sharing data using standards like the GJXDM and regional data sharing, as well as deploying field-based technology at a steady rate. Agencies are also more aware of outside technology advances thanks to programs that encourage technology transfer from the private sectors like the CRM industry. However, it is time to start pulling these resources together to move from counting problems to solving them.
One public example was developed by the Richmond (Va.) Police Department in 2003. Richmond had issues with random gunfire every New Year's Eve holiday. The police department took several steps to solve this problem using predictive analytics. Historical data was analyzed and prioritized with considerations made for locations that had escalations in gunfire complaints since the first of the year. The information provided by the crime analysis unit included a plan to strategically place personnel where analysts expected history to repeat itself during the 2003 to 2004 New Year's holiday. The outcomes included a 49-percent reduction in gunfire complaints on New Year's Eve, and a 26-percent reduction in gunfire complaints on the following days. Forty-five weapons were also seized. An unanticipated additional benefit to this operation was a $15,000 reduction in overtime expenses, as personnel were placed where they were needed most and 50 people were able to take the holiday off, according to McCue.
But the success of those agencies using predictive analysis is not always known, she cautions. Those agencies pursuing advanced analytics in the law enforcement arena are not always making their information public. Until that happens, she says it's difficult to know with any certainty who is making progress and what outcomes are actually occurring.
Another real challenge is that uninformed consumers can wreak havoc on the very tools that are necessary to the predictive process. Sen. Russ Feingold of Wisconsin, for instance, nearly killed advanced analytics for police agencies when he introduced the Data Mining Moratorium Act of 2003. The bill died in committee, but it's a reminder that law enforcement must be careful as it proceeds with data gathering in the name of solving problems.
A regional approach
The trend for local agencies to combine their resources to form regional approaches to predictive crime analysis is also helpful for many reasons. First and foremost, the costs are spread among several agencies, allowing them to maximize contributed dollars. The open sharing of data also increases the potential for agencies to place less emphasis on their political boundaries, since criminals do not tend to observe those same boundaries when they commit crimes. An added benefit to combining data is that rare and unusual events are more likely to be captured and considered when predicting crimes for an area.
As police problem solving strives to move from a reactive (counting mode) to a proactive (anticipation and response mode), advanced predictive analytics offers the opportunity to change outcomes. While this is a simple statement, the implications have far-reaching consequences and positive potential for any law enforcement agency. A police agency willing to commit to the concept and pull resources together to put predictive analytics into practice is an agency that will be far ahead of those agencies still counting their crimes but not doing anything helpful with the numbers.
Problem solving has always been an expectation of law enforcement agencies, but today the industry is at a point where the power of prevention has significant and perhaps immeasurably positive consequences. It appears with predictive crime analysis that an ounce of prevention is definitely worth a pound of cure.
Eric Mills is the commander of the Strategic Services Division in the Pasadena (Calif.) Police Department.