March Machine Learning Mania

README

This repository contains code for training various machine learning models to predict tournament results using logistic regression, random forest, and gradient boosting techniques. The implementation supports both men's and women's tournaments.

Model Training

Logistic Regression

Steps:

1. Mount Google Drive to access datasets.
2. Load and preprocess data.
3. Train a logistic regression model using LogisticRegression from sklearn.
4. Calibrate the model using CalibratedClassifierCV.
5. Generate predictions for tournament teams.
6. Save results in a submission file.

Code Overview:

• Loads data from Google Drive.
• Performs feature engineering (calculates score difference as a feature).
• Splits data into training and testing sets.
• Uses a pipeline with standardization and logistic regression.
• Calibrates the model for better probability estimates.
• Generates win probability predictions for each team.
• Creates a CSV submission file for tournament predictions.

Random Forest

Steps:

1. Load preprocessed features from logistic regression section.
2. Train a RandomForestClassifier.
3. Calibrate using CalibratedClassifierCV.
4. Predict tournament outcomes.

Issues:

• NameError: name 'X' is not defined missing feature definitions.

Gradient Boosting Machines (LightGBM)

Steps:

1. Define a LGBMClassifier model.
2. Calibrate with CalibratedClassifierCV.
3. Train and predict probabilities for tournament teams.

Issues:

• NameError: name 'X_train' is not defined due to missing variable assignment.

Grid Search CV (LightGBM)

Steps:

1. Define a parameter grid for hyperparameter tuning.
2. Use GridSearchCV with brier_score_loss to optimize model performance.

Submission File Generation

A function create_submission_file_2025() generates a CSV file containing predictions for matchups in the 2025 NCAA tournament. It combines results from men's and women's tournaments.