2025 Wharton Data Science Competition

NCAA Basketball Projections Machine Learning System

This project was built for the Wharton High School Data Science Competition. It utilises a Random Forest Classifier to predict the winning percentages of teams in the NCAA basketball East region. A model is trained on the game data provided and generates predictions along with two visualisations.

Features

• Predicts home team winning probabilities for the specific matches, outputting them in a CSV
• Generates a bar chart to visualise the predictions
• Ranks teams in North, South and West regions by expected wins against every other team in the region
• Exports a pdf of a sample decision tree used in the Random Forest model. Note: The Random Forest model uses many decision trees; the pdf generated is of one of them

Prerequisites

• Python 3.7+
• Required Python libraries:
  • pandas
  • scikit-learn
  • matplot lib
  • seaborn
  • graphviz (ensure graphviz is located in System PATH)

Install dependencies using:

pip install pandas scikit-learn matplotlib seaborn graphviz

Data Preparation

Input Files

The following files, given by the Wharton competition, are used:

1. Team Region Groups.csv: Contains team-region mappings.
2. games_2022.csv: Historical game statistics from the 2022 season.
3. East Regional Games to predict.csv: Matchups to predict (East Regional games).

Data Processing

• Merges team region data with game statistics
• Processes numeric features, given by the competition, and training the model with them
• Combines home/away game data into a single dataset with feature prefixes to abide by conventions in files given by the competition

Usage

1. Run the script:

python main.py

Outputs

• East_Predictions.csv: Contains predicted winning percentages for home teams.
• decision_tree.pdf: Visualisation of one decision tree from the Random Forest.
• A bar plot displaying sorted winning percentages (shown interactively).
• North_Ranking.csv, South_Ranking.csv and West_Ranking.csv: Files containing all teams in a given region ranked in descending order by their expected wins

Model details

• Trained with several values given by the competition
• Random Forest Classifier with default parameters

Team: Markham Masters, Perú