TRAP: A Predictive Framework for Trail Running Assessment of Performance

Abstract

Trail running is an endurance sport in which athletes face severe physical challenges. Due to the growing number of participants, the organization of limited staff, equipment, and medical support in these races now plays a key role. Monitoring runner's performance is a difficult task that requires knowledge of the terrain and of the runner's ability. In the past, choices were solely based on the organizers' experience without reliance on data. However, this approach is neither scalable nor transferable. Instead, we propose a firm statistical methodology to perform this task, both before and during the race. Our proposed framework, Trail Running Assessment of Performance (TRAP), studies (1) the the assessment of the runner's ability to reach the next checkpoint, (2) the prediction of the runner's expected passage time at the next checkpoint, and (3) corresponding prediction intervals for the passage time. To obtain data on the ability of runners, we introduce a Python package, ScrapITRA, to access the race history of runners from the International Trail Running Association (ITRA). We apply our methodology, using the ITRA data along with checkpoint and terrain-level information, to the "holy grail" of ultra-trail running, the Ultra-Trail du Mont-Blanc (UTMB) race, demonstrating the predictive power of our methodology.