A network-driven methodology for sports ranking and prediction

Recent years have seen increasing interest in ranking elite athletes and teams in professional sports leagues, and in predicting the outcomes of games. In this work, we draw an analogy between this problem and one in the field of search engine optimization, namely, that of ranking webpages on the Internet. Motivated by the famous PageRank algorithm, our TeamRank methods define directed graphs of sports teams based on the observed outcomes of individual games, and use these networks to infer the importance of teams that determines their rankings. In evaluating these methods on data from recent seasons in the National Football League (NFL) and National Basketball Association (NBA), we find that they can predict the outcomes of games with up to 70% accuracy, and that they provide useful rankings of teams that cluster by league divisions. We also propose some extensions to TeamRank that consider overall team win records and shifts in momentum over time.