For some agencies, the notion of predictive policing conjures images that are more science fiction than actual police work.

Proponents and developers of predictive policing technology say their solutions don’t look to supplant normal police operations but to enhance them. Generally speaking, predictive policing looks to predict the locations of crimes, offenders and their identities, and potential crime victims.

Law enforcement users from Philadelphia, Pennsylvania, to Los Angeles, California, have reported good outcomes using predictive policing software, and several studies point to its effectiveness.

According to a 2015 study published in the Journal of the American Statistical Association, predictive policing solutions do not necessarily lead to more arrests, but they can have a preventive effect that, when combined with effective police work on the ground, can lead to lower crime rates. Police patrols using a certain type of software-driven forecasting tool led to an average 7.4 percent reduction in crime.

“From an ethnographic perspective, we believe predictive policing operates at a local level through short-term disruption of criminal opportunities,” the study authors wrote in the report.

A representative scenario arises when an officer shows up at a location designated as high risk. An offender who lives or works in that area sees the officer and decides to lay low or even run. In that time they are laying low, they are in no position to commit a crime. If the offender comes out a few hours later and again sees the officer in the same or a nearby hot spot, the deterrence effect may last well beyond those particular policing events.¹

Jeff Brantingham is one of the authors of the study. He’s also the co-founder and chief of research and development at PredPol, a Santa Cruz, California, firm widely regarded as a leader in predictive policing. An early supporter of this technology as an effective law enforcement tool, Brantingham said the tool continues to advance and may be more relevant to law enforcement than ever.

PredPol defines its mission as “the practice of identifying the times and locations where specific crimes are most likely to occur, then patrolling those areas to prevent those crimes from occurring.”² This speaks directly to a conviction that predictive policing works best when it functions alongside an officer’s daily work, rather than subsuming or supplanting it.

“Algorithms can do a great job of predicting when and where crimes are most likely to occur,” Brantingham said. “The second part of that is that no matter how sophisticated your algorithm is, or the nature of your computer interface, policing is a person-centric exercise and interacting with communities and solving problems. Nothing will change that fundamental activity.”³

Law Enforcement and Community

Predictive policing is a bona fide trend in law enforcement, particularly for larger departments. However, there is also no dispute that it has the potential to raise questions about privacy.

“Big Data” is now a common term around the world, as is the concept of mining data for any number of purposes. That phenomenon naturally brings other issues along with it, particularly in a law enforcement context. For example, mining social media data could have the potential to “predict” potential criminals who in reality have only a passing social media connection to established perpetrators.

This can not only raise red flags for any citizen concerned about individual privacy, but also lead to concerns by officers and departments striving to improve relations with their communities.

Each agency determines whether or how to respond to such concerns, but one option is transparency. Transparency in data is a standard practice of Tyler Technologies, a Plano, Texas, company that creates a suite of solutions for use across the public sector, from schools to permitting departments.

For law enforcement, Tyler assists with predictive policing by leveraging freely available records information from law enforcement, fire and rescue, and other public agencies and pooling it together to form a more complete picture of criminal activity in a given area.

“When we look at predictive policing, we try to do more data-driven policing info and get it out to folks in command centers, so they can make better decisions,” said Russell Gainford, vice president of product development for Tyler’s public safety division.⁴

A central tool in this area for Tyler is Socrata Public Safety Analytics, a map-centric tool that provides a detailed picture of crime, accidents, tickets, law enforcement and fire incidents, and calls for service by displaying data in an easy-to-use interface for quick trend analysis.

As an example, Tyler officials point to a recent collaboration between the fire and police departments in Topeka, Kansas. Compiling and sharing data, officers determined that the fires were started intentionally, then used commonalities between the fires to build a profile of a potential suspect and to identify areas where the next fire might occur. Investigators surveilled the identified areas and subsequently apprehended the suspect.

This type of predictive policing not only is effective as a law enforcement tool, but also can enhance community relations through transparency.

“It’s a citizen-facing tool,” Gainford said. “We started on the citizen side, so you can instantly respond to requests from the public.”

The data is easy to view and manipulate, with the resulting models helping to protect police from unfounded accusations from the community.

“You can create heat maps over certain time periods or geo fences,” Gainford said. “When you come into the solution, it aggregates data very quickly. What it’s providing you is logistics, and providing insights [and] honeycomb maps that can really be used to justify use of force.”

Machine Learning

“Machine learning” is a subset of the wider concept generally known as artificial intelligence. Machine learning is a method of data analysis that automates analytical model building. Specially designed software systems can gather and “learn” from the analysis, identifying patterns with minimal human intervention.

Machine learning is the technology underlying many predictive policing solutions, including PredPol.

“There is a lot of interest in machine learning because it can help accomplish tasks because it had to do [it] in any other way. PredPol has the latest machine learning and mathematical modeling,” Brantingham said. “PredPol is a software service company doing real-time crime forecasting in the field. It’s continuously ingesting information on where and when crimes are occurring and figuring out when new hot spots are going to appear. New predictions are always available to officers.”

Although these systems are incredibly complex, they are not especially daunting to officers and agencies because they operate much like regular technology that is familiar to most web and mobile device users.

“There is a lot of machine learning, but the officer doesn’t see that,” Brantingham said. “They just see the models. It’s a lot like apps on mobile devices. There is a lot of math going on in the background of every phone app, but you don’t see it. It’s streamlining the way you use time and resources.”

The solution itself is seamless, Brantingham said, with new predictions arriving each morning. Mapping is fairly targeted. “The target areas are only 500 feet by 500 feet,” Brantingham said. “So, this is more than just identifying a neighborhood. This is more like a block, a micro-scale area on the ground.”

New data arrives regularly, but it is provided in a way that allows agencies to work with the data when and how it makes the most sense for their schedules and duties.

“The predictions are delivered to the watch commander in an email inbox every morning, and it’s a fresh set of predictions based on data,” Brantingham said. “You can add it right into roll call and briefings, with hot sheets that map it out, or you can see it on tablets or in the car. When they’re not on another call, they’re putting themselves into these locations.”