Evolutionary Madness!: Applying an Evolutionary Algorithm to Optimize Team Selections for an NCAA Tournament Contest

• Activity Number: 43
• Type: Contributed
• Date/Time: Sunday, July 29, 2012 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistics in Sports
• Author(s): Douglas Noe*+ and Geng Chen
• Companies: Miami University and The University of Texas Health Science Center at Houston
• Keywords: evolutionary algorithm ; optimization ; cost constraint ; NCAA ; basketball ; tournament

Abstract

As one of the nation's most popular sporting events, the NCAA tournament always attracts millions of people's attention. In a typical "bracket" contest, people participate by filling in tournament brackets with the goal of correctly predicting the most round- and often upset-weighted game winners. We examine a unique contest structure in which each entrant selects a collection of teams subject to a seed-based cost constraint with the simple goal of maximizing the unweighted total number of wins.

We focus on one objective among a variety of team-selection strategies. Given a prior set of beliefs about the outcome probabilities for each potential tournament match, we seek to determine the set of teams with the greatest probability of achieving a preselected threshold number of wins. Due to the complex nature of the sample space, this problem is not easy to solve analytically. Instead, we apply an evolutionary algorithm to heuristically search the sample space by mimicking the process of natural evolution.

We examine the success of our approach using data from each tournament from 2003 through 2012 and validate our algorithm using actual tournament results.