preview of "Predictive policing" publication
mullerj 0100000HUM Visits (6701)
Here is a preview of an article I wrote on Predictive Policing which compares IBM SPSS to other software solutions. Please ask permission before referencing anything in this blog post - as this draft is currently being reviewed by the Orange County Sheriff's Office for their Crime Analyst Training & Mentoring Program (CAMP) newsletter.
John K. Muller, Ph.D.
Software Engineer, IBM (current)
Orange County Sheriff’s Office (1/2011 - 6/2011)
What is predictive policing?
What is predictive policing? Does predictive policing overlap with any of the following concepts: smart policing, predictive analytics, police analytics, crime forecasting, risk-based policing, intelligence led policing, problem oriented policing, or proactive policing? Unfortunately, there is not a generally agreed upon definition of predictive policing. For the purposes of this article predictive policing involves using computer models, crime and data from the physical environment, to predict crimes, anticipate criminal activity and prevent it.1 Predictive policing is about taking into consideration multiple factors to predict future crime, versus only considering where past crimes have occurred.
Unfortunately it is impossible to provide a step-by-step guide for implementing predictive policing, since each department has unique goals, objectives, resources, tools, data sources and crime types of most interest. All of these factors will help decide which predictive policing software solution, if any, is most appropriate. Recent advances in computing power and software capabilities warrants, at the very least, consideration of existing software solutions that may be of benefit to crime analysts and sworn personnel.
Following is a high level review of a few predictive policing software solutions. While a software solution may help enhance one’s ability to support and engage in predictive policing, the crime analyst will always play a central role. For example, law enforcement expertise is required to help setup models, interpret results, and decide on what action to take based on those results.
GeoEye Analytics Signature Analyst
Analytics Signature Analyst provides an empirical, statistical approach which
focuses on social, cultural, and physical variables to produce hot-spot maps. These maps are used to indicate areas where
future events are likely to occur. If
you are a National Geos
Risk Terrain Modeling (RTM)
Risk terrain modeling provides a way to measure multiple risk factors across spatial units. The resulting product produces a “risk terrain” map which displays the intensity of all risk factors for each spatial unit. This map shows where conditions are favorable for crimes to occur. If you currently are using ArcMap with the Spatial Analyst extension, then you already have all the tools you need to get started with RTM. A free reference book is also available which provides a step by step guide.2 The website also provides documented known risk factors by crime type to help identify aggravating and mitigating risk factors. You can decide which risk factors to include through testing with statistical analysis the place-based correlation of each risk factor with the outcome event of interest. An ad hoc approach can also be used, but this requires that you justify your decisions based on existing theory, empirical research and professional experience. At each spatial unit, every risk factor can have only one of two possible values (0 or 1 for aggravating factors, -1 or 0 for mitigating factors). Multiple case studies, and documents on applying RTM to specific crime types (e.g., shootings, aggravated assault) can be found online.
Information Builders’ Law Enforcement Analytics (LEA)
Analytics includes web forms that law enforcement officers in the field can use
to help them predict the probability of crimes occurring within a particular
patrol area. Their website provides
additional information, including case studies with the Char
IBM SPSS Statistics
predictive analytics was adopted by the Memphis
police department, as part of their Blue CRUSH (Criminal Reduction Using
Statistical History) project. Blue CRUSH
is an impressive case study, showing a 30% reduction in serious crime, and a 15%
reduction in violent crime. The IBM SPSS
solution makes use of ArcGIS for the mapping component, but other mapping tools
should work as well. Part of the reason
SPSS has had such notable success may be in part to the “Cha
When conducting your own research be sure to not limit yourself to only the software tools mentioned here but investigate other tools and approaches for predictive policing that fit well with your department’s needs. It is difficult to provide a comprehensive review for many of these solutions as the algorithms used are often proprietary. In addition, the variability of these approaches makes comparisons between them impossible without hands on experience with the software. In fact, an enterprising crime analyst could employ multiple predictive policing software tools to conduct a head-to-head comparison using real-world data, and publish the results so others in the field could benefit from the findings. The results of such a study will not be as definitive as stating that one solution is better than another, but it is more likely the study would conclude that they compliment each other in very specific ways. For example, for a particular agency, one solution may yield better predictive results for a certain type of crime, whereas a different solution is better in other situations.
When evaluating solutions keep in mind that a more complicated predictive model does not necessarily produce better predictions. In fact, each variable must be grounded in theory before it is included in the model. Identifying these variables correctly is critical, therefore adequate time should be spent conducting literature searches and expert interviews. During this process, always keep in mind that any measurement of success must focus on whether or not the final product helps prevent and reduce crime.
“… the true promise of crime mapping lies in its ability to identify early warning signs across time and space, and inform a proactive approach to police problem solving and crime prevention.” - Groff and La Vigne (2002)4
The resulting product of a predictive solution will vary by department. For one law enforcement agency the solution might be centered around setting a threshold alert, for example, when a certain number of auto burglaries occur in a given area an alert is sent out with the goal of fixing a developing issue before it becomes a new hotspot. For a different department, the solution might involve forecasting specific types of crime based on things like weather, the outcome of local sporting events, the age of neighborhoods, etc. Whatever form the resulting product takes, it is important to be able to measure the impact/success of predictive policing efforts.5
1. Define your objectives first. How does your department want to use predictive policing? What do you want to get out of it? Are you interested in predicting certain types of crimes?
2. Take an inventory of your departments current resources, data, and outstanding requirements.
3. Conduct your own detailed investigation of possible solutions and expand your search beyond just the solutions covered here. A good starting point is the IACA website6 which includes the “First Predictive Policing Symposium Report” from 2009.
4. Ask for personalized sales/marketing pitches for the solutions that seem most likely to fit your situation.
5. Know beforehand how you are going to measure success, at the very least your model should predict crime better then a model based solely on last period’s criminal activity.
6. Since your model is not going to always be correct, be sure to manage expectations across your entire organization.
1 Susan C. Smith (2011). Predictive Policing. Shawnee, Kansas Police Department.
2 Caplan, J. M. & Kennedy, L. W.
(2010). Risk Terrain Modeling Manual: Theoretical Framework and Technical Steps
of Spatial Risk Assessment. Newark, NJ: Rutgers
Center on Public
Security. (available for download at http
3 Directions Magazine (2011). Police Departments Reduce Crime with Spatial Analytics. Webinar.
4 Groff, E. R., and La Vigne, N. G., (2002).Forecasting the Future of Predictive Crime Mapping. In Analysis for Crime Prevention, Crime Prevention Studies, Volume 13, Nick Tilly (Ed), Lynne Rienner Publishers.
5 Personal communication with Susan C. Smith, BS, MBA, CLEA, Crime Analyst, Shawnee Police.
Note: The opinions stated in this article are those of the author and do not necessarily represent the views of the Orange County Sheriff’s Office or IBM.