Abstracting Glicko-2 for Team Games
Garrick J. Williams · OhioLink ETD Center (Ohio Library and Information Network) · 2013
Research experiments were conducted to further the development of rating player skill within the context of online team games with small team sizes.Modern rating systems tend to display inferior performance when individually measuring players within a team of six or less members.This thesis proposes three modifications to rating update algorithms that abstract a match between two teams as multiple one-on-one matches.These update methods were designed for adaptation with existing skill rating systems based on Glicko or Glicko2.Using a robot soccer simulation environment, an automated testing suite was developed to show how these methods affect the behavior of Glicko2.The results of an experiment utilizing this suite show the viability of these methods and paint a picture for future research towards improving rating system performance for team games with small team sizes.4.3 Ranking vs. Win Percentage 4.4 Convergence V.