Analysis and Augmentation of Ranking Systems: A Case Study

Abstract

The proliferation of online matchmaking and the continued development of gambling in regard to sports makes it more important now than ever to be able to identify the skill levels of players and the likelihood of particular match outcomes. Many good systems use neural networks and support vector machines to predict match outcomes, but these methods are computationally expensive, and it is often impossible or just plain inefficient to compute updated rankings with these systems. We draw our attention to the most popular and computationally inexpensive skill updates systems of chess and online matchmaking: Elo and Microsoft’s TrueSkill. We offer a much needed analysis of how these systems perform in a case study on predicting tennis match outcomes. We analyze the effectiveness of these systems in different situations, and we develop some augmentations and test their effectiveness.