Analysis of the 72-h ultramarathon using a predictive XG Boost model

Abstract Background Ultramarathon running enjoys unwavering popularity. This includes the 72-h run, the longest time-limited ultramarathon based on hours and not days, yet this specific race format remains understudied. In particular, we are still determining where the fastest 72-h ultra-marathoners originate or where the fastest races are held. The aim of the present study was to investigate the origins of the best performers and the locations of the fastest races. Methods A machine learning model based on the XG Boost algorithm was built to predict running speed based on the athlete´s gender, age group, country of origin, the country where the race was held, the kind of race course (road, trail, track), and the elevation (flat, hilly). Model explainability tools were then used to investigate how each independent variable would influence the predicted running speed. Results A total of 2,857 race records from 1,870 unique runners from 36 different countries participating in 55 races held in 22 countries between 1989 and 2022 were analyzed. Athletes from the USA account for more than 2/3 of the sample size. Also, more than 3/4 of the participants competed in USA-based races. Athletes from Ireland, Japan, and Ukraine were the fastest. In respect of the fastest races, they were held in Ukraine, The Netherlands, and Japan. The model rated the country of event as the most important predictor followed by the race characteristics of elevation and race course, athlete country of origin, age group, and gender . On average , men were 0.33 km/h faster than women. The fastest running speeds were achieved by runners in age group 45–49 years. Conclusions The country of the event was found to be the most important predictor in the 72-h run. Despite the dominance of runners from USA and the predominance of courses in the USA in terms of participation, athletes from Ireland, Japan, and Ukraine achieved the fastest times, while Ukraine, The Netherlands, and Japan were found to host the fastest courses.