Predicting and Understanding Australian Rules Football Using Markov Processes

Abstract

Using Markov processes Australian Football matches were analysed by modelling the flow of the play, between zones on the field and match events such as possessions, stoppages and scores. Transition matrices were created according to the probabilities of transitioning between zones and match events. Match simulations were conducted using these transition matrices for the purposes of both pre-match predictions and in-play predictions, and run iteratively to attain a large statistical sample. Home ground advantage was also applied. Through this process a model was created which was found to have similar distributions of match events as actual Australian Football matches. The model predicted the result of Australian Football matches with similar accuracy to other non-Markov models, but with the added ability to understand Australian Football. Several coaching and analytical applications are presented based on the Markov model, including quantifying a team’s game style, tracking the effect of adjustments to game style, and identifying the most effective course of action under certain circumstances. Potential applications of this model include an in-game decision support tool for coaches, enhanced broadcast media, and simulation of match tactics for clubs outside of game day.