Machine Learning Analysis of Horse-Rider Dyad Performance in Show Jumping

This study utilized machine learning techniques to analyze factors that contribute to the performance of horse and rider combinations (dyads) in Fédération Équestre Internationale (FEI) jumping as ranked by the FEI Combination in Jumping ranking list. Data on horses and riders were gathered from 6000 FEI webpages, and machine learning models were applied on both a priori data such as sire and color and a posteriori data such as competition statistics to study not only the factors that contribute to the success of a horse and rider combination but also to examine if any of these already known factors can provide insight into future performance. Data were analyzed using the CatBoost algorithm and Shapley Additive Explanations (SHAP). All machine learning models were trained and evaluated with K-fold cross validation on these different combinations of factors to produce a statistically significant R
²
value from a sample in a regression of the predicted rankings and the actual rankings, with the model trained on all data leading to a R
²
of 0.67. These results vastly outperformed regression analysis of individual variables against the target. Analysis of important features revealed that the features that contribute most to the models’ predictions are those related to the amount of practice and experience of both the horse and rider.