Enhanced Prediction of March Mania Using Conference and Historical Information

• Activity Number: 193 - Contributed Poster Presentations: Section on Statistics in Sports
• Type: Contributed
• Date/Time: Monday, July 30, 2018 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Sports
• Author(s): Guanhong Miao* and Samuel Wu
• Companies: and University of Florida
• Keywords: NCAA Basketball; Bradley-Terry Model; Kaggle Competition

Abstract

Bradley-Terry model is commonly used for pairwise comparison outcome. In this paper, we propose to improve prediction of men's NCAA basketball tournament outcome by including conference structure, performances from previous seasons, ranking of recruitment classes, and time trend during the regular season. In the enhanced Bradley-Terry model, the strength parameter at the end of regular season is allowed to be different from that of the beginning of the season for each team. And the strength parameters for teams in the same conference are expected to be similar. Strength parameters are estimated based on adjusted log-likelihood with penalty function on variations of strengths within conference and strength differences between beginning and end of season. The proposed method is evaluated based on the NCAA basketball tournament data in the past 33 years.