This study aimed to compare the predictive accuracy of absolute and relative external load indices (ELI) across three machine learning models, and predict the rating of perceived exertion (RPE) of elite Gaelic football players using ELI, personal characteristics, wellness scores, and training workloads. ELI and related variables were collected from 49 elite Gaelic football players over three competitive seasons resulting in 1617 observations. ELI included total distance, high speed running distance (≥ 4.72 m · s-1), and number of accelerations and decelerations (n ± 3 m · s-2), expressed in both absolute and relative terms. Variables related to personal characteristics, wellness scores, and training workloads were also included. Data were analysed using decision tree, random forest (RF), and bootstrap aggregation (BS) models. The RF model had the highest predictive accuracy using absolute and relative ELI only, at 54.3% and 48.3%, respectively. Total and relative distance were the strongest predictors of RPE in the RF model, accounting for 38.8% and 27.9% of the normalised importance. The BS model had the highest accuracy at 67.0% and 65.2% for absolute and relative ELI when performed in conjunction with the related variables, respectively. The current models demonstrate potential to predict RPE and subsequently optimise training load in Gaelic football.
Read full abstract