| 07-Nov-2018

Using Machine Learning to Improve Safety

When Matei, currently an M4 (Grade 10) student, first applied to UTS, he and his family were inspired by the school’s strong academic reputation. And before Matei had even taken his first class, he found reassurance that UTS was where he belonged.

“My first interaction with other UTS students was at the F1 orientation night,” Matei recalls. “Right then, it clicked with me that this was where I could have meaningful conversations with people who understood and enjoyed the same kinds of things as me.”

One of the things Matei has enjoyed in his years at UTS is the opportunity to gain experience and skills in machine learning. And when a moment of inspiration presented itself during a trip to Washington, D.C. this past summer, he came up with an idea of how to put these skills to use for society.

“In Washington, they’ve implemented a program where you can pick up a scooter and drop it off wherever you want to, and you pay a fee for however long you use it,” Matei says. “One night I noticed a scooter that was lying on its side, completely destroyed.”

“This was a sore spot for me because I know how hard it is to build something that works, and I wondered who would do such a thing.”

Matei began researching publically-available data from Toronto’s police force about the location and time of day of crimes in the city, with the idea that programs like Washington’s scooter service could use the information to suggest and incentivize users to leave the devices in safe spaces.

The initial goal became unachievable as he dug deeper into the data, “The information that was available didn't contain reliable statistics about this new type of ride-share-like transportation. Current data referenced much broader vandalism crimes. This would have created an inaccurate predictions model that had no definitive way to predict when and where a scooter was most likely to be vandalized.“

However, this didn’t stop Matei from realizing that his project had another application in predicting much broader types of crime in a certain region.

Since the end of the summer, Matei has been working on a program that would have applications for both individual citizens and organizations. “There are common ideas about where it is and isn’t safe to walk around after dark, but what I found in my research is that a lot of it isn’t true. There were a lot of surprises in what my program predicted.”

Matei built the program by using a sub-field of machine learning known as unsupervised learning, which involves a computer identifying patterns in data without human help. Using an unsupervised learning algorithm known as K Means, Matei’s model is able to classify the available information into distinct subgroups based on factors including the date, location, and type of crime. Inputting new data such as a new location or a different date, the algorithm would generate predictions that include the probability of a crime occurring then and there, as well as the similarity of that crime to other crimes that the model has seen before (vandalism, assault, etc.).

Matei also envisions the Toronto PD, ride-hailing services such as Uber, and public transit commissions incorporating his program. “Toronto police can use this tool to be more proactive in fighting crime by patrolling different areas of Toronto at different times. Uber can find safe dropoff and pickup locations based on the time of day and location of certain rides, and the TTC can help people stay safe by allowing riders to get off in between stops in places that the program predicts are dangerous.”

Matei will continue to refine his program as part of his year-long Civics project.

“We were asked to develop something that makes some kind of social change,” Matei says. “I thought I could use this project, as opposed to a one-time volunteering opportunity, to build something that makes an impact and can be used well into the future.”