This project aims to estimate realistic and interpretable win probabilities for horse races by modelling horse speed as a continuous target, rather than directly predicting race outcomes. It employs a classic Ordinary Least Squares (OLS) regression, chosen for its transparency and robust statistical assumptions, followed by Monte Carlo simulations to convert predicted speeds into valid race-level win probabilities.
The model is particularly useful in settings where market odds are unavailable or unreliable and prioritises interpretability and replicability over black-box complexity.
| File Name | Description |
|---|---|
trainData.csv |
Training dataset containing race features and speed targets. |
testData.csv |
Test dataset used for final model evaluation and probabilistic predictions. |
predicted_probabilities.csv |
Model-generated win probabilities for each horse in each test race. |
Horse Racing Probabilistic Modelling Report_Yi-Lung(Dragon) Tsai.pdf |
Detailed technical report explaining methodology, modelling, and evaluation. |
Horse Racing Probabilistic Modeling.ipynb |
Full Jupyter Notebook including code for data preprocessing, model fitting, and probability generation. |