Khoury News

Trouble finding a parking spot? Students and faculty develop app to predict availability

For many big-city dwellers, the search for parking is one of immense frustration. Now, armed with data and machine learning tools, a team of student and faculty researchers believe they have the beginnings of a solution.

December 4, 2005

by Laila Griffin

Share article

Close this dialog window

For Sarita Singh, parking can take up to 20 minutes. And sometimes, she never finds a spot. 

“Having lived all my life in urban cities, parking has always been a big challenge,” said Singh, an associate teaching professor at Khoury College in Seattle. “Wherever we go, we have to park our cars somewhere.”  

She is not alone. Studies indicate that almost 30% of urban traffic consists of drivers who are simply searching for parking. So Singh and Mario Nascimento, Khoury College’s Vancouver-based director of Pacific Northwest research, tasked Vancouver-based graduate students Huijia Wang, Ke Wang, Tianhao Zhang, and Yuming Sun with solving the problem as their capstone project. They had 13 weeks to do so.  

The result: a predictive parking mobile app that integrates historical and real-time data to help drivers find available parking spaces. The app currently works for Singapore, which has made the necessary data publicly available. Source code for the proposed app is also publicly available. 

“I like solving problems,” Nascimento said. “This is a very interesting problem because it’s very pragmatic. It’s real life. Students can relate to it.” 

Existing apps such as Google Maps can locate and display parking lots but lack data on parking space availability. These apps can also encourage people to use specific parking lots that pay for more advertising.  

For the student team seeking a better alternative, the first challenge, and one of the toughest, was finding accessible datasets and application programming interfaces (APIs), tools that allow software systems to exchange data.  

“It was important to ensure that all the resources were available,” said Singh, who lived in Singapore for 17 years and still calls the city home. “We built it for Singapore because they had the data for it. They [had] the real-time API, the real-time parking availability, and real-time weather conditions in addition to the historical data.”  

Using Singapore’s open data portal to access information across multiple parking lots, the team created a dataset to feed into the model, which they combined with historical weather data. Precipitation, they argued, was particularly important to observe because it can alter driving and parking patterns. For instance, people might cancel or postpone driving trips due to rainy weather.  

The team selected a long short-term memory (LSTM) machine learning model, which is well-suited to work with time-series data to forecast trends and outcomes based on historical information.  

“Using the past, we can try to learn patterns so it can predict the future,” Nascimento said. “If an interesting fact or pattern emerges, the model keeps this in its memory.”  

Surprisingly, precipitation did not play as big a role as anticipated. The team suspected that the limited variation in Singapore’s weather patterns played a role, as Nascimento can attest. 

“I lived in Singapore for six months and I remember almost every day, at about 4 p.m., a cloud of rain was coming,” he said. 

Still, precipitation data proved useful.  

“Incorporating precipitation data into predictive models generally provided a slight improvement in the accuracy of parking availability predictions,” the team noted. 

Throughout the project, Nascimento and Singh encouraged students to be problem-solvers.  

“Part of the research process is to learn,” Nascimento said. “But we don’t just throw them in the swimming pool. We say, ‘Here’s how we’re going to swim.’” 

In the future, the students hope to extend the project by observing factors such as parking payment options, traffic conditions, and dynamic pricing, where parking rates are adjusted based on factors like time of day or location.  

“We could then follow the same approach, but we could add different features and contextualize it to different locations,” Singh said. 

Singh and Nascimento also advocate for the collection and release of more comprehensive data to improve predictive models like theirs.  

“It’s going to require investment infrastructure,” Singh said. “That means there has to be some sensor that captures this information and makes it publicly available to people. That would be an important step if we want this kind of application to be more widespread.”  

Ultimately, the application could assist with urban planning, lessen traffic congestion, and decrease unnecessary carbon emissions — all while empowering users to make more informed decisions. The very usage of the app would generate more of this data, according to Nascimento. 

“There is a lot of opportunity for authorities to take this data and plan their city infrastructure better,” Singh said. “I think that there is lots and lots of potential for this particular application.”