Prediction of game performance in Australian football using heart rate variability measures

In-match player performance, measured by data from Geographical Positioning System (GPS) devices, was predicted with a correlation coefficient of greater than 0.7. Predictions were based on heart rate variability measures and used advanced regression techniques based on machine learning. These techniques included methods for the selection of variables to be included in the regression study. Results indicate that variable selection using a wrapper subset method with a genetic algorithm outperformed both principal components analysis and the default method of using all variables. The success of prediction of match performance suggests a potential for new tools to assist the team coach in player selection and management of player training. This work also provides the possibility for a training programme to be adjusted specifically to meet the challenges of the size of the playing field and the temperature likely to be encountered on the day of the match.