Christopher M Young1, Wei Luo2, Paul Gastin3, Jacqueline Tran4, Dan B Dwyer3. 1. Centre for Sport Research, Deakin University, Geelong, Australia. Electronic address: yochr@deakin.edu.au. 2. School of Information Technology, Deakin University, Geelong, Australia. 3. Centre for Sport Research, Deakin University, Geelong, Australia. 4. Centre for Sport Research, Deakin University, Geelong, Australia; Geelong Cats Football Club, Geelong, Australia.
Abstract
OBJECTIVES: To identify novel insights about performance in Australian Football (AF), by modelling the relationships between player actions and match outcomes. This study extends and improves on previous studies by utilising a wider range of performance indicators (PIs) and a longer time frame for the development of predictive models. DESIGN: Observational. METHODS: Ninety-one team PIs from the 2001 to 2016 Australian Football League seasons were used as independent variables. The categorical Win-Loss and continuous Score Margin match outcome measures were used as dependent variables. Decision tree and Generalised Linear Models were created to describe the relationships between the values of the PIs and match outcome. RESULTS: Decision tree models predicted Win-Loss and Score Margin with up to 88.9% and 70.3% accuracy, respectively. The Generalised Linear Models predicted Score Margin to within 6.8 points (RMSE) and Win-Loss with up to 95.1% accuracy. The PIs that are most predictive of match outcome include; Turnovers Forced score, Inside 50s per shot, Metres Gained and Time in Possession, all in their relative (to opposition) form. The decision trees illustrate how combinations of the values of these PIs are associated with match outcome, and they indicate target values for these PIs. CONCLUSIONS: This work used a wider range of PIs and more historical data than previous reports and consequently demonstrated higher prediction accuracies and additional insights about important indicators of performance. The methods used in this work can be implemented by other sport analysts to generate further insights that support the strategic decision-making processes of coaches.
OBJECTIVES: To identify novel insights about performance in Australian Football (AF), by modelling the relationships between player actions and match outcomes. This study extends and improves on previous studies by utilising a wider range of performance indicators (PIs) and a longer time frame for the development of predictive models. DESIGN: Observational. METHODS: Ninety-one team PIs from the 2001 to 2016 Australian Football League seasons were used as independent variables. The categorical Win-Loss and continuous Score Margin match outcome measures were used as dependent variables. Decision tree and Generalised Linear Models were created to describe the relationships between the values of the PIs and match outcome. RESULTS: Decision tree models predicted Win-Loss and Score Margin with up to 88.9% and 70.3% accuracy, respectively. The Generalised Linear Models predicted Score Margin to within 6.8 points (RMSE) and Win-Loss with up to 95.1% accuracy. The PIs that are most predictive of match outcome include; Turnovers Forced score, Inside 50s per shot, Metres Gained and Time in Possession, all in their relative (to opposition) form. The decision trees illustrate how combinations of the values of these PIs are associated with match outcome, and they indicate target values for these PIs. CONCLUSIONS: This work used a wider range of PIs and more historical data than previous reports and consequently demonstrated higher prediction accuracies and additional insights about important indicators of performance. The methods used in this work can be implemented by other sport analysts to generate further insights that support the strategic decision-making processes of coaches.