An optimized ratings-based model for forecasting Australian Rules football

Abstract

Building a ratings model for forecasting the success of a sporting team requires the careful consideration of many factors, such as the home ground advantage and opponent quality. In this research, we build an optimized Elo ratings model for forecasting Australian Rules football (AFL), which incorporates the home ground advantage (ground familiarity and travel fatigue) and seasonal decay (initial ratings); ratings are then updated between games based on the difference between the expected and actual margins of victory. Match information gathered from the 2000 and 2001 seasons was used as a training set for the forward prediction of the 2002 to 2009 seasons. The model is then evaluated based on the number of predicted winners, the Average Absolute margin of Error (AAE) and the Return on Investment (ROI).