2025 SPECIAL EDITION: Featuring the Winning Team from the 2025 Wharton High School Data Science Competition

AUTHORS

Author: Shivam Gupta, ‘26, Lambert High School, GA
Advisor: Stephanie Beaulieu, Lambert High School, GA

ABSTRACT

We present an ensemble approach to NCAA women’s basketball prediction, combining four independent models to produce robust regional team rankings and win probabilities. Our primary method was an Elo rating system with basketball-specific considerations including home-court advantage, rest days, and travel distance. We verified its accuracy and reliability by cross-validating with three additional models: Dean Oliver’s Four Factors, temporal logistic regression, and a network-based PageRank model.

Using over 5,300 Division I games, our models provided a consensus, exhibiting the Elo rating system as the most conservative and reliable. Although our predicted rankings and win probabilities solely relied on our modified Elo rating model, the other three played a crucial role informing our final decision with data and statistics.