Salisbury builds new computer model to predict crimnes

By Katie Redefer
Posted 9/28/22

Salisbury is set to launch a recently developed predictive policing project in which artificial-intelligence software can forecast when and where crimes might occur.

The city’s Information …

You must be a member to read this story.

Join our family of readers for as little as $5.99 per month and support local, unbiased journalism.

Already a member? Log in to continue.   Otherwise, follow the link below to join.

Please log in to continue

Log in

Salisbury builds new computer model to predict crimnes


Salisbury is set to launch a recently developed predictive policing project in which artificial-intelligence software can forecast when and where crimes might occur.

The city’s Information Services department set out to develop a deep-learning algorithm in 2018 after it received a grant from the Governor’s Office of Crime Control & Prevention with the goal of analyzing local crime patterns to predict law enforcement needs specific to the city.

According to Mayor Jake Day, this modeling tool allows police officers to more strategically place themselves in areas where crime is predicted to occur, and thus can respond to crimes faster – especially when it comes to crimes such as motor vehicle theft, robberies and assault. 

“Our Information Services department created an AI-based tool that had 97.4 percent accuracy in predicting where a crime would occur,” the mayor said.  “There is evidence to suggest an officer being present, whether on foot patrol or parked, may reduce the likelihood that crime happens.”

After receiving the state grant, city data scientist Grace Malcom completed a unique deep-learning algorithm in 2019 – one which well exceeded the industry-standard of predictive policing.

The modeling was then sent to state officials for further evaluation and research, and later passed off to research teams at the University of Maryland. 

Salisbury Police Chief Barbara Duncan said the emphasis placed on analyzing crime patterns on the municipal level – not just national or regional – is crucial to the model’s accuracy. 

“Initially, we were interested in reducing crime by using data to get ahead of crime through preventive efforts,” Duncan said. “Once we experienced success with commercially available predictive models, we began to look at whether predictive analytics might produce better outcomes if the analytics included non-traditional data sets which were more connected to the specific needs of our community.

“We realized that the best way forward would be to begin working on deep machine learning at the local level,” she said. 

Last month, the University of Maryland research team, led by Ted Knight, presented the city with a report outlining its progress over the past year in further developing the model to local and state officials.​​

The team will continue adding various regional and state-level data in order to improve the accuracy of the model by “considering diverse sets of activity, law enforcement, behavioral health, and health and environmental data,” according to Day. 

“A major goal of this project is to minimize and remove any biases or positive/negative feedback loops – which are generally associated with these models – while maintaining a high degree of accuracy and functionality within the platform,” said John O’Brien, Assistant Director of Information Services. “Developing this model in an open source environment with active input from diverse but collaborative organizations is the best way to achieve this goal.”

Baltimore City began implementing a similar predictive policing program in 2018, and subsequently saw a reduction in homicides and shootings. In Annapolis, however, the City Council there voted against funding for predictive modeling in 2020, with councilmembers expressing concerns the model could lead to increased racial profiling and discrimination.

Day maintains that predictive modeling is incremental in protecting the city’s residents and enforcing its laws. 

“We believe that using predictive modeling at the municipal level is one of the most effective tools we have at our disposal to reduce crime in our city,” Day said. “There is extensive evidence that predictive modeling is accurate and effective, and we are proud to be one of the key players at the forefront of these developments in our state.”

Members and subscribers make this story possible.
You can help support non-partisan, community journalism.