March Madness Machine Learning 🏀

Overview

The following codebase predicts matchups between NCAA men's and women's basketball teams, using logistic regression trained on historical game data.

The results of the algorithm's output for each matchup are used to solve both the men's and women's NCAA tournament bracket.

The solution CSV files are compliant with Kaggle's March Machine Learning Mania challenge.

Tools Used

• sqlite

• pandas

• scikit-learn

• bracketeer

Running the Model

First, install all Python requirements: pip3 install -r requirements.txt . Next, visit the Kaggle March Machine Learning Mania page to download the provided test/training data.

Add these files to a directory titled march-machine-learning-mania-2023 in the root directory of the project.

From here, configure the run_config.json file as desired:

{
  "sport": "women",
  "execution_name": "V1",
  "add_external_sources": false,
  "run_data_setup": false,
  "aggregate_team_data": false,
  "transform_and_test": false,
  "export_content": true
}

The configuration options are described below:

sport: defines the context for the model (men or women) - it will use the corresponding data as specified here

execution_name: name given to your execution, allowing for multiple executions of different names

add_external_sources: use the external data in the external_sources directory for the model

run_data_setup: create/set up the SQLite database (this can be skipped if you've already run this step and your data has not changed)

aggregate_team_data: this will likely remain true unless previously run - aggregates season data for each team to provide season average stats

transform_and_test: create transformed data from the raw stats to be used in training, testing and predictions by the model

export_content: export the results to CSV and use bracketeer to create an image of bracket predictions

Once configured, run the model: python3 main.py

Women's Bracket 2023 Using Bracketeer: