Improving sport horse selection using machine learning: Early prediction of elite-level show jumping performance
- 1Department of Veterinary Sciences, University of Turin, Grugliasco, Italy
- 2Department of Veterinary Science, University of Parma, Parma, Italy
Abstract
Early identification of sport potential in show jumping horses remains challenging because competition outcomes vary widely, and elite-level performance develops after considerable time. In this study, we evaluated the applicability of machine learning approaches to competition records from horses identified through FEI participation, complemented by national-level records from Sweden, Belgium and France, with the aim of identifying horses likely to reach elite performance. Performance data from Fédération Équestre Internationale (FEI) competitions were analysed using an Extreme Gradient Boosting (XGBoost) classifier trained on longitudinal competition histories recorded before horses reached ten years of age. Animals were classified as champions if they achieved at least one clear round at 160 cm, while non-champions were defined as those that completed their competitive career without reaching this level. Model performance was assessed using nested cross-validation and compared with two single-rule-based classifiers based on maximum obstacle height thresholds. XGBoost outperformed baseline approaches across all evaluation metrics, achieving an average area under the ROC curve of 83.5% and an average precision of 80.0%. Analysis of probability trajectories showed that predictive information accumulates progressively over a horse’s career, with increasing separation between champion and non-champion predictions as competition history expands. Model performance improved substantially from approximately seven years of age onward. Permutation importance and SHapley Additive exPlanations (SHAP) analyses revealed that obstacle difficulty, performance consistency and age-related progression were among the most influential predictors. Overall, these findings demonstrate that interpretable machine learning models can extract meaningful predictive signals from early competition trajectories and support evidence-based talent identification in show jumping, while maintaining attention to performance development and horse welfare.
Related
- Hit Selection Using SSMD-Based Machine Learning Performance Metrics in High-Throughput Screening Assays
- CaliPPer: quantifying, predicting and improving AI model performance for binding prediction
- Improving the Usability of Donor Data from Registries Using an Extreme Learning Machine Approach to Upgrade Low-/Mid- to High-Resolution HLA Data
- Regional climate variability and rice yields prediction in Bangladesh using machine learning and deep learning models
- Integrating Multisource Target Evidence with Machine-Learning Potency Prediction for Alzheimer's Disease Drug Discovery
BibTeX
@article{zanchi2026improving,
title = {{Improving sport horse selection using machine learning: Early prediction of elite-level show jumping performance}},
author = {Marco Zanchi and Clara Bordin and Michela Ablondi and Vittoria Asti and Andrea Summer and Emanuela Valle and Laura Ozella},
journal = {PLOS One},
year = {2026},
month = sep,
doi = {10.1371/journal.pone.0357799},
url = {https://doi.org/10.1371/journal.pone.0357799},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - Improving sport horse selection using machine learning: Early prediction of elite-level show jumping performance
AU - Marco Zanchi
AU - Clara Bordin
AU - Michela Ablondi
AU - Vittoria Asti
AU - Andrea Summer
AU - Emanuela Valle
AU - Laura Ozella
PY - 2026
DA - 2026/09/15
JO - PLOS One
DO - 10.1371/journal.pone.0357799
UR - https://doi.org/10.1371/journal.pone.0357799
AB - Early identification of sport potential in show jumping horses remains challenging because competition outcomes vary widely, and elite-level performance develops after considerable time. In this study, we evaluated the applicability of machine learning approaches to competition records from horses identified through FEI participation, complemented by national-level records from Sweden, Belgium and France, with the aim of identifying horses likely to reach elite performance. Performance data from Fédération Équestre Internationale (FEI) competitions were analysed using an Extreme Gradient Boosting (XGBoost) classifier trained on longitudinal competition histories recorded before horses reached ten years of age. Animals were classified as champions if they achieved at least one clear round at 160 cm, while non-champions were defined as those that completed their competitive career without reaching this level. Model performance was assessed using nested cross-validation and compared with two single-rule-based classifiers based on maximum obstacle height thresholds. XGBoost outperformed baseline approaches across all evaluation metrics, achieving an average area under the ROC curve of 83.5% and an average precision of 80.0%. Analysis of probability trajectories showed that predictive information accumulates progressively over a horse’s career, with increasing separation between champion and non-champion predictions as competition history expands. Model performance improved substantially from approximately seven years of age onward. Permutation importance and SHapley Additive exPlanations (SHAP) analyses revealed that obstacle difficulty, performance consistency and age-related progression were among the most influential predictors. Overall, these findings demonstrate that interpretable machine learning models can extract meaningful predictive signals from early competition trajectories and support evidence-based talent identification in show jumping, while maintaining attention to performance development and horse welfare.
SN - 1932-6203
ER -