Machine learning predictions from University of Guelph researchers are helping Canadians get a better picture of what they will need to spend on groceries next year.
The 2025 Canada Food Price Report, now in its 15th edition, relies on a team of U of G researchers in food science, food economics, global food sustainability and artificial intelligence to inform food price predictions for the next year.
The 2025 report predicts food prices will rise by three to five per cent next year, meaning an average family of four should expect to spend $16,833.67 on food in 2025, an increase of $801.56 from 2024. The highest increases are expected in meat and vegetables.
Dr. Graham Taylor, professor in the School of Engineering and lead researcher of the Machine Learning Research Group, has helped forecast food prices for the report over the past six years.
“Our team uses machine learning to build models that look at historical food price data and other factors that can then forecast what food prices will be in the next year,” he says.
Food prices prediction driven by AI
Taylor and his team, who work in partnership with the Vector Institute, used large language models (LLMs) for the first time to generate forecasts for the 2025 report. This year, PhD candidate Kristina Kupferschmidt acted as the technical lead, proposing novel ways of leveraging LLMs for forecasting as part of her thesis research.
The LLMs simulated subject-matter experts, examining variables that might be important for each food category included in the report and then performing forecasts based on those variables.
“There are so many factors driving food prices, like global conflicts, exchange rates and climate change,” Taylor says. “LLMs can read the whole internet and unlock that knowledge to guide our predictions, potentially offering a better forecast than a person or traditional statistical model.”
When building models each year, Taylor’s team looks back on past reports to determine which technique will provide the most accurate forecasts. They place the models into an internal competition, comparing actual outcomes forecasted by past reports to each model’s predictions.
“The model that performs the best for each type of food is used in our forecast,” Taylor says. “LLMs were the winning models in several categories, so we used them to predict 2025 food prices in those categories.”
Optimizing price prediction with machine learning
The LLM-based methodology the team used this year was contributed by Dr. James Requeima, a postdoctoral fellow at the Vector Institute, who collaborated with Kupferschmidt to adapt the approach to food prices.
Besides LLM-based forecasts, the team considered several families of models: Time series foundation models, transformers specialized to forecasting, as well as traditional statistical and deep learning approaches.
As food prices rise and Canadians continue to deal with a cost-of-living crisis, Taylor says accurate predictions are more important than ever.
“Information from the Canada Food Price Report could help people optimize their food budget,” Taylor suggests. “This might include buying non-perishable items when they’re on sale or looking for lower-priced alternatives.”
Taylor notes that while price increases have stabilized, costs will likely never drop back to pre-pandemic levels.
“Those gains were locked in and now we’re making smaller gains on top of that,” he says. “That drives home the pain that people are still feeling. The prices don’t go down; the increases just become smaller.”
Undergrad students gain valuable experience
This year, two undergraduate co-op students from the College of Engineering and Physical Sciences, Zohrah Bee Varsally and Mya Simpson, joined the team and played a vital role in building the predictions for the report. Their participation was funded by the Doody Family Chair Undergraduate Research Assistant Fund for Women in Engineering, a fund aimed at encouraging more women to enter and graduate from engineering programs.
Varsally and Simpson are both in biomedical engineering but developed their machine learning skills by working on the Canada Food Price Report.
“It was awesome to get experience through hands-on coding and to be given the trust and encouragement to contribute to this report,” Simpson says.
The students also had the opportunity to make major contributions to a report that will be read by millions of Canadians.
“Researching food prices gave us a strong sense of what’s happening in the world and how that affects people’s everyday lives,” Varsally says.
Adds Simpson: “It was exciting to be part of work that’s actively going to impact how people prepare for the future.”
The Canada Food Price Report is produced collaboratively by Dalhousie University, University of Guelph, University of British Columbia and University of Saskatchewan. Other contributors from U of G include Ontario Agricultural College food economists, engineers and team members from the Arrell Food Institute.