Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Food for thought: How can machine learning help better predict and understand changes in food prices?

Kristina L. Kupferschmidt, James Requiema, Mya Simpson, Zohrah Varsallay, Ethan Jackson, Cody Kupferschmidt, Sara El-Shawa, Graham W. Taylor

TL;DR

The paper tackles predicting Canadian food price inflation and understanding how context and data curation affect forecast accuracy in CFPR 2025. It proposes a data-centric, human-in-the-loop forecasting framework combining domain-expert data curation and LLM-assisted curation across multiple model families (statistical, DL, Transformer, foundation, and LLMs). Key results show that climate and geopolitical regressors often yield the strongest gains, ensembles of top-performing models outperform baselines, and LLM inputs can improve or harm stability depending on prompting and input design. The work demonstrates practical improvements for forecasting Canada's food prices and outlines directions for integrating context-aware prompting and routing strategies to further enhance accuracy.

Abstract

In this work, we address a lack of systematic understanding of fluctuations in food affordability in Canada. Canada's Food Price Report (CPFR) is an annual publication that predicts food inflation over the next calendar year. The published predictions are a collaborative effort between forecasting teams that each employ their own approach at Canadian Universities: Dalhousie University, the University of British Columbia, the University of Saskatchewan, and the University of Guelph/Vector Institute. While the University of Guelph/Vector Institute forecasting team has leveraged machine learning (ML) in previous reports, the most recent editions (2024--2025) have also included a human-in-the-loop approach. For the 2025 report, this focus was expanded to evaluate several different data-centric approaches to improve forecast accuracy. In this study, we evaluate how different types of forecasting models perform when estimating food price fluctuations. We also examine the sensitivity of models that curate time series data representing key factors in food pricing.

Food for thought: How can machine learning help better predict and understand changes in food prices?

TL;DR

The paper tackles predicting Canadian food price inflation and understanding how context and data curation affect forecast accuracy in CFPR 2025. It proposes a data-centric, human-in-the-loop forecasting framework combining domain-expert data curation and LLM-assisted curation across multiple model families (statistical, DL, Transformer, foundation, and LLMs). Key results show that climate and geopolitical regressors often yield the strongest gains, ensembles of top-performing models outperform baselines, and LLM inputs can improve or harm stability depending on prompting and input design. The work demonstrates practical improvements for forecasting Canada's food prices and outlines directions for integrating context-aware prompting and routing strategies to further enhance accuracy.

Abstract

In this work, we address a lack of systematic understanding of fluctuations in food affordability in Canada. Canada's Food Price Report (CPFR) is an annual publication that predicts food inflation over the next calendar year. The published predictions are a collaborative effort between forecasting teams that each employ their own approach at Canadian Universities: Dalhousie University, the University of British Columbia, the University of Saskatchewan, and the University of Guelph/Vector Institute. While the University of Guelph/Vector Institute forecasting team has leveraged machine learning (ML) in previous reports, the most recent editions (2024--2025) have also included a human-in-the-loop approach. For the 2025 report, this focus was expanded to evaluate several different data-centric approaches to improve forecast accuracy. In this study, we evaluate how different types of forecasting models perform when estimating food price fluctuations. We also examine the sensitivity of models that curate time series data representing key factors in food pricing.

Paper Structure

This paper contains 23 sections, 1 equation, 5 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Objectives
  3. Related Works
  4. Human-centric forecasting
  5. Incorporating context into forecasting
  6. Methodology
  7. Target task and dataset -- Predicting food-related consumer price index
  8. Data curation to provide specific context
  9. Domain expert data curation
  10. LLM-persona data curation
  11. Additional Inputs for LLM Forecasters
  12. Models
  13. Time-series forecasting models
  14. LLMs as forecasting models
  15. Model Evaluation and Final Forecast Generation
  16. ...and 8 more sections

Figures (5)

  • Figure 1: Comparison of citation trends across different domains of foundation models.
  • Figure 2: Performance changes for different context inclusion techniques -- Averaged over all food categories and model types
  • Figure 3: Model rankings for each food category in CPI sorted in order of increasing complexity of model classes. Results represent single best-performing regressors from each model. Models are sorted in order of capacity increasing from left to right with model classes being statistical, deep-learning-based, transformer-based, foundation models, and LLMs. Model capacity was estimated based on best estimates of # of parameters.
  • Figure 4: Forecasts and projected year over year (2025/2024) % change for the food CPI categories included in Canada's Food Price Report
  • Figure 5: Inclusion of specific time-series in evaluated groupings