A machine learning AI used to predict the winners and under/overs of NBA games. Takes all team data from the 2007-08 season to current season, matched with odds of those games, using a neural network to predict winning bets for today's games. Achieves ~69% accuracy on money lines and ~55% on under/overs historical anylsis. Outputs expected value for teams money lines to provide better insight. The fraction of your bankroll to bet based on the Kelly Criterion is also outputted. Note that a popular, less risky approach is to bet 50% of the stake recommended by the Kelly Criterion.
Use Python 3.11. In particular the packages/libraries used are...
- Tensorflow - Machine learning library
- XGBoost - Gradient boosting framework
- Numpy - Package for scientific computing in Python
- Pandas - Data manipulation and analysis
- Colorama - Color text output
- Tqdm - Progress bars
- Requests - Http library
- Scikit_learn - Machine learning library
Make sure all packages above are installed.
$ git clone https://github.com/kyleskom/NBA-Machine-Learning-Sports-Betting.git
$ cd NBA-Machine-Learning-Sports-Betting
$ pip3 install -r requirements.txt
$ python3 main.py -xgb -odds=fanduel
Odds data will be automatically fetched from sbrodds if the -odds option is provided with a sportsbook. Options include: fanduel, draftkings, betmgm, pointsbet, caesars, wynn, bet_rivers_ny
If -odds
is not given, enter the under/over and odds for today's games manually after starting the script.
Optionally, you can add '-kc' as a command line argument to see the recommended fraction of your bankroll to wager based on the model's edge
This repo also includes a small Flask application to help view the data from this tool in the browser. To run it:
cd Flask
flask --debug run
# Create dataset with the latest data for 2023-24 season
cd src/Process-Data
python -m Get_Data
python -m Get_Odds_Data
python -m Create_Games
# Train models
cd ../Train-Models
python -m XGBoost_Model_ML
python -m XGBoost_Model_UO
All contributions welcomed and encouraged.
Please be advised this repository is provided as an analytical tool and should be used as such. The maintainers are not responsible for the results or accuracy of the models generated by this tool. Historical performance and model evaluation does not guarantee future outcomes. You must exercise your own judgment in evaluating future events. Users should independently assess and verify any information obtained from this tool before making decisions based on such information.