RESUMO
Player movement in rugby league is complex, being spatiotemporal and multifaceted. Modeling this complexity to provide robust measures of player activity and load has proved difficult, with important aspects of player movement yet to be considered. These include the influence of time-varying covariates on player activity and the combination of different dimensions of player movement. Few studies have simultaneously categorized player activity into different activity states and investigated factors influencing the transition between states, or compared player activity and load profiles between matches and training. This study applied hidden Markov models (HMMs)-a data-driven, multivariate approach-to rugby league training and match GPS data to i) demonstrate how HMMs can combine multiple variables in a data-driven way to effectively categorize player movement states, ii) investigate the influence of two time-varying covariates, score difference and elapsed match time on player activity states, and iii) compare player activity and load profiles within and between training and match modalities. HMMs were fitted to player GPS, accelerometer and heart rate data of one English Super League team across 60 training sessions and 35 matches. Distinct activity states were detected for both matches and training, with transitions between states in matches influenced by score difference and elapsed time and clear differences in activity and load profiles between training and matches. HMMs can model the complexity of player movement to effectively profile player activity and load in rugby league and have the potential to facilitate new research across several sports.
We successfully derived player activity and load profiles in both training and match contexts in a data-driven and multivariate way using hidden Markov models.HMMs can be used to investigate the probability of changing between activity states as a function of time-varying covariates, augmenting current activity profiling practice.We discovered key differences between the activity and load profiles between training and matches in rugby league. In particular, a very directed high-speed running state in training that is seldom accessed by players in matches.We demonstrated how visualizing the output of HMMs can provide decision support by facilitating comparisons of activity and load profiles within and between players in matches and training.We posit that the methodology detailed in this paper can become a standardized approach to player activity and load profiling based on player movement data across multiple sports because it is flexible, data-driven, multivariate and statistically robust.