Abstract

As sports technology evolves, the technical area of player management is rapidly changing along with artificial intelligence (AI). Monitoring the physical and psychological status of players is essential. Predicting the rate of perceived exertion (RPE) is a common practice as it captures self-reported training intensity. However, by excluding coaches’ training plans and relying only on players’ data, existing studies lack an external benchmark for intended intensity, hindering calibrated interpretation of subjective RPE and actionable decision-making. They also depend on device-based data collected during training imposing spatiotemporal constraints. Hence, this study proposes a new indicator integrating data from both coaches and players, collected via an app-based approach, to summarize training behaviors. Using the rate of intended exertion (RIE) to represent coaches’ intended intensity, we defined the difference between RIE and RPE (DIP) as a new indicator reflecting discrepancies between coaches’ intent and players’ perception during training. We constructed a task for predicting the indicator. For the prediction, athlete self-report measures (ASRM) were used. The architecture of the prediction model includes a transformer to leverage the self-attention mechanism capturing temporal characteristics of ASRM. Experimental results indicated a mean absolute error (MAE) of 0.6791, outperforming similar existing studies. This study makes three major contributions. First, we propose a simple and powerful indicator reflecting both coaches’ intent and players’ perception. Second, our model uses app-based data minimizing spatiotemporal constraints. Third, unlike previous studies relying on device-based data collected during training, our model provides predictions in advance using only ASRM accumulated prior to training.