Imputation of non-participated race results

Most current solutions in cycling analytics focus on one specific race or participant, while a sports-wide system could render huge benefits of scale, by automating certain processes. The development of such a system is, however, heavily inflicted by the large number of non-participations as most riders do not compete in all races. Therefore, value imputation is required. Most popular value imputation techniques are developed for cases where part of the data is fully observed, which is not the case for cycling race results. While some methods are adapted to situations without complete cases, this is not the case for the cross-sectional imputation algorithm suggested by multiple previous studies (i.e., KNN imputation). We therefore suggest an adaptation to the KNN imputation algorithm which uses expert knowledge on race similarity in order to facilitate the deployment of the algorithm in situations without complete cases. The method is shown to be the most performant predictive model and does this within a competitive computation time.

ISBN:9783031020445

Publication year:2022

Accessibility:Open