Predicting Crime with Big Data
Thursday, March 7, 2013 at 12:05PM
Dr. Jennifer Bachner is Program Coordinator for the MA in Government program. Her work on predictive policing is supported by the IBM Center for the Business of Government.The rise of big data is shifting decision-making practices in all sectors of society. Journalists like David Brooks observe that “data-ism” is the “rising philosophy of the day” and businesses like McKinsey and IBM have focused their efforts on developing products and services that harness the power of big data. The big data revolution means that organizations of all types will need to collect, clean, analyze and act upon increasingly massive amounts of quantitative information to remain competitive.
Among those on the frontier of this paradigm shift are law enforcement agencies. Through what is often referred to as “predictive policing,” police departments are experiencing unprecedented success using data and analytics. Intelligence-led policing (i.e. the adoption of CompStat) emerged in the 1990s and greatly improved accountability by tracking information such as crime and arrest rates. Predictive policing builds upon this foundation. By examining patterns in past crime data, in conjunction with environmental characteristics, analysts can generate amazingly accurate forecasts about where crime is likely to occur. Officers are then deployed according to these forecasts.
The Santa Cruz Police Department, for example, partnered with social scientists at UCLA and Santa Clara University to develop software that assigns the probability of crime occurring to 150 by 150 meter cells on a map. Prior to their shifts, officers are notified of the 15 cells with the highest probabilities, and during their shifts, they can log into a web-based system to access updated, real-time probabilities. While officers are encouraged to view the maps as one of many tools in their crime prevention kits, many of those who integrate the information into their on-the-ground decision making have experienced marked declines in crime rates on their beats.
The applications of predictive policing extend well beyond mapping locations with increased likelihoods of crime (“hot spots”). The Baltimore Police Department, for example, has used predictive methods to inform its offender interdiction tactics. With a serial robber, analysts can use analytics to pinpoint the likely location of the suspect. To accomplish this, analysts first employ an iterative algorithm to calculate the center of minimum distance (CMD) between crime scenes, which is assumed to be the offender’s residence. An analysis of all possible routes from the CMD to the crime scenes and back again frequently reveals a limited number streets and times the offender uses. Police can then conduct an efficient stakeout and apprehend the suspect.
Social network analysis has proven to be another effective prediction tool. Analysts with the Richmond Police Department recently used this type of analysis to identify central (mathematically speaking) members in a homicide suspect’s social network. Police had been searching for the suspect for over a month. A few days after police notified key members in the network of their search, the suspect turned himself in. The police had successfully shut down the suspect’s social resources and, with no safe haven, he submitted himself to the authorities.
More and more police departments across the country are implementing predictive policing programs as the technology and training become more accessible. Further, the increasing computing power and data storage capacities available to police departments allow analysts to integrate more information into predictive analyses. Over the past few decades, criminologists have identified numerous characteristics associated with heightened criminal activity, including the availability of escape routes (e.g. highways and bridges), presence of adult retail establishments, weather patterns, payday schedules, times of day, days of the week and even moon phases. Through collaborative efforts, social scientists, crime analysts and police officers are discovering new ways to leverage this information and translate it into actionable recommendations that prevent crime.
We can also expect predictive policing to improve as information sharing becomes easier. Serial criminals often cross jurisdictional boundaries. This presents a problem for crime analysts, as the accuracy of predictions is positively correlated with information completeness. Recognizing this challenge, the federal government has supported the development of the Law Enforcement Information Exchange (LInX), which serves as a data warehouse for all participating police agencies. Individual agencies populate the database with their crime data, which can then be accessed by other agencies. Other information-sharing systems, such as Digital Information Gateway (DIG), is likewise making data analysis, visualization and interpretation easier and more accurate.
Law enforcement agencies are certainly not the only organizations benefiting from predictive analytics. Retail companies, intelligence agencies, financial institutions and marketing firms are just a few of the organizations using predictive methods and big data to improve their efficiency and success rates. And this trend is likely to continue. IBM estimates that “90% of the data in the world today has been created in the last two years.” This is a great time for undergraduate and graduate students to focus their academic careers in the field of data and analytics.