This project aims to predict college football game results during conference play (Week 4 and beyond). Predictions are likely to reflect real results when the deviation from the spread is greater than 3 points. Predictions are made weekly using the make_predictions.py script, which leverages helper files and data files.
make_predictions.py: Main script for making weekly predictions.- Helper Files:
select_features.py: Selects the relevant features for the model.update_game_data.py: Updates game data used for predictions.
- Data Files:
XGBoost_for_spread_cfb.dat: Pre-trained XGBoost model (included for completeness, but not used--I found that the neural net was more accurate on its own in a validation set).cfb_feature_normalizations.dat: Normalization parameters for features.features_for_cfb_model.dat: Feature set used by the model.neural_net_for_spread_cfb.dat: Pre-trained neural network model.
- Training File
College Football Model Fitting.ipynb: Shows the process of training the models.
Making Predictions: Run make_predictions.py to generate predictions for the upcoming week's games. The script will output the predicted spreads and the deviation from the actual spreads. This uses live dates so works after week 3 of a season.