How About Crime Preventers Instead of Crime Stoppers?
Kris Andreychuk, Supervisor of Community Safety in the City of Edmonton’s Neighbourhood Empowerment Team, was frustrated.
His teams of social workers, police officers and youth workers were great at helping move criminals or juvenile offenders onto a better path after they’d been involved in a crime, but he couldn’t help thinking:
Wouldn’t it be better if we could predict areas where crime would be most likely to be brewing and work to improve those neighbourhoods before crimes happen?
That led him to contact Stephane Contre, Chief Analytics Officer of Edmonton’s Open City Team, to see how they could use the masses of data the City had access to in order to identify patterns that are likely to cause crime in a particular area.
Not the Usual Suspects
In our guts, most of us think we have a pretty good idea of where crime is likely to occur and the sorts of people who are likely to commit it. And, sure, certain parts of any city are likely to have higher crime rates than others.
But even in a slum, the vast majority of residents are not criminals. So how do you narrow it down to a much more specific area — say a couple of city blocks — so you can be more targeted in your interventions?
That’s what they did in this project.
From 16 Crime Variables to 233?
Starting with a literature review, Andreychuk and Contre came up with a list of 16 variables that research showed were typically associated with crime. But none of them told the whole story.
So they reached out to the community and asked citizens, social workers, police and others to list variables they thought might be somehow correlated with crime, either with higher crime rates or with lower ones.
They came up with a list of 233 possible factors.
The great thing about artificial intelligence and machine learning, Andreychuk and Contre told me in our interview after their recent talk to the Marketing Research & Intelligence Association’s Alberta Chapter, is that “we didn’t have to be the goalkeeper,” deciding whose ideas to include and whose to exclude.
The system could analyse them all.
Bring On The Data Sets!
Then they looked at what data sets they could get a hold of that might include the 233 variables the public and experts came up with. This included data from sources as diverse as Parks and Recreation databases to police crime stats.
They ended up looking at variables that, on their own, would seem to have no obvious correlation, such as the location of picnic tables.
In the end, they determined that no single variable had a strong correlation with crime (not even things you’d think would) but there were 92 distinct combinations of factors that were predictive of very specific areas where crime was a higher risk.
So, for example, a combination of recovered stolen vehicles, a low density of picnic tables, and a high number of youth centres, was one such pattern.
But Data Doesn’t Tell The Whole Story
As powerful as the data is, you still need to bring the experience of humans on the front line to bear in order to come to useful conclusions about the data.
Adding more picnic tables won’t solve the problem.
In one case, it turned out that the recovered stolen vehicles weren’t actually from auto thefts by locals, but rather from “dial-a-dopers” in another part of the city, who’d leave their stolen vehicles there. This was well known to the local youth workers, but not something that the artificial intelligence would have put together.
The Data’s In, But How Do You Change Beliefs?
In the interview we also talk about how to change people’s long-held beliefs about the causes of crime and ways to fight it. And that comes down to many of the change management factors we’ve discussed often in this blog.
As Contre put it,
“The analytics is the easy part…. The hard part is that other part of changing mentalities.”