Key result
Data-centric AI decision tree models achieve ~80% AUC for predicting non-contact injuries versus baseline.
Why the study?
Preventive measures to reduce harmful injury episodes are needed because injuries impact team performance and the sports industry's economic situation.
Machine learning models utilizing GPS workload data and data-centric AI techniques can predict non-contact injuries in professional soccer players with moderate to high discrimination.
May aid workload monitoring in soccer; hypothesis-generating pending prospective validation.
One big concern in soccer professional teams is to search for preventive measures to reduce the frequency of harmful episodes in their athletes since these episodes greatly impact the sports industry and affect both the team’s performance and the association’s economic situation. Thus, the present work proposes a methodology to predict non-contact injury episodes that may affect them in a microcycle through Data-centric AI concepts. The prediction model is trained using a dataset related to professional soccer athletes. The most interesting result were with AUC-ROC of 79,8%. About the performance improvement strategies applied, the best undersampling ratio was 70/30, PCA with one or two principal components did best, and the Decision Tree algorithm excelled.
No takes yet. Share an insight, caveat, or question.
Melo et al. (2024) studied Non-contact traumatic injuries (n=79). Data-Centric AI predictive modeling (Decision Tree with PCA and undersampling) vs. Baseline predictive models without data-centric techniques was evaluated on Area Under the ROC Curve (AUC-ROC) for injury prediction. A Decision Tree predictive model utilizing data-centric AI techniques, including a 70/30 undersampling ratio and PCA, achieved an AUC-ROC of 79.8% for predicting non-contact injuries.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: