Examination of player role in the Australian Football League using match performance data

ABSTRACT This study developed multiple methods to determine player role in Australian Rules football utilising objective match performance data. Specifically, Australian Football League (AFL) Player Ratings from the 2016 AFL season were used to classify players into seven a priori determined playing roles, as well as determine levels of individual player similarity. Mean values for the 11 AFL Player Ratings categories were calculated for each individual player, and a performance profile created based on the relative contribution of points from each category to that players overall rating total. A decision tree model incorporated five of the 11 categories to classify player role at an accuracy of 74.3% (95% confidence interval = 70.5–77.9% across 10-fold cross-validation). Role classification was most accurate for key forwards, midfielders and general defenders, whilst the midfield-forward role was most difficult to define objectively. A Euclidean distance measure was used to determine the most similar pairs of individual players within the AFL, as well as from an intra-club perspective. An application was also developed to visually represent the similarity of players within the squad of a single AFL club. Sporting organisations may apply the methods provided here to support decisions regarding player selection and recruitment.