349 – Speed Session #3: Topics in Methods and Applications, Part 2
In Pursuit of Perfection: An Ensemble Method for Predicting March Madness Match-Up Probabilities
Sara Stoudt
Smith College
Loren Santana
Smith College
Ben S. Baumer
Smith College
Warren Buffett's 1 billion dollar prize for the perfect 2013 - 2014 NCAA bracket and Kaggle's March Machine Learning Madness competition increased the stakes of predicting this year's tournament results. Adding to the core data given by the Kaggle competition of team wins and loses since 1995 - 1996, we gather other team data to allow for more flexibility and options for modeling. We investigate different ensemble methods in R that contain strategies such as a modified PageRank algorithm, decision trees, neural networks, among others to predict probabilities of different match-ups and create a bracket for this season. We discuss strengths and weaknesses of our approach.