A team of University of Toronto students has won two out of three awards at the 2023 CANSSI Ontario French Trot Horse Racing Forecasting Competition: the Performance Award and the Innovation Award. The team included PhD students Cynthia Yip and Juliette Zaccour, both from the Faculty of Information. Juliette Zaccour is now affiliated with the Oxford Internet Institute at the University of Oxford.
As winners of the Performance and Innovation Awards, the team received a combined prize of $8,500.
Cynthia and Juliette presented a robust, iterative solution for predicting race outcomes, emphasizing data cleaning, feature engineering, and modeling. Their approach involved refining variables, imputing missing values, and crafting insightful features such as race seasonality, relative horse age, preferred surface type, rest period, and past performance metrics for horses, jockeys, and trainers.
Preprocessing steps accounted for both categorical and numerical variables, addressed missing values in engineered features, and made other minor adjustments. The XGBoost algorithm was selected for its balance of performance and efficiency. They focused on optimizing log loss while maintaining transparency through interpretable features.
Model evaluation demonstrated high accuracy (89.3%), though precision and recall were more modest. The model correctly predicted the winner for 631 out of 2,140 races—approximately 29.5% of races. Their solution emphasized real-world adaptability and included a dedicated notebook for testing the model on new datasets.
About the Competition
The French Trot Horse Racing: Forecasting Competition, held from September 22 to November 7, 2023, was open to undergraduate and graduate students and postdoctoral fellows, who were challenged to build the most effective models for predicting outcomes in French Trot Horse Racing. The event provided a platform for participants to showcase their expertise in machine learning and statistics.
The Performance Award recognized the team with the most accurate winner forecasts over a three-month validation dataset.
The Innovation Award celebrated creativity, highlighting the team that explored innovative concepts, introduced novel features, or shared interesting stories.
Learn more about the 2023 CANSSI Ontario Forecasting Competition here