Trustworthy AI for multimodal load-response modelling in team sports: a scoping review and full-text validation audit
Year: 2026
Type: article
Abstract: Introduction Artificial intelligence (AI), machine-learning and conventional statistical prediction methods are increasingly used in team-sport athlete monitoring, but it remains unclear whether current studies are validated and reported in ways that support trustworthy prospective use. This scoping review mapped predictive-modelling studies in team-sport athlete monitoring and multimodal load-response modelling and audited their validation and reporting practices. Methods Searches were conducted in OpenAlex, PubMed, Scopus and SPORTDiscus for records published from 2010 to 2026. After deduplication, screening and updated OpenAlex rerun screening, 258 studies were included and audited from full text. Systematic duplicate screening was not performed; full-text eligibility assessment and original audit coding were completed by a single reviewer. Data were extracted on publication characteristics, sport context, data modalities, modelled outcomes, model families and validation practices. A 10-domain trustworthy-AI audit assessed prediction context, temporal validity, repeated-observation independence, external validation, leakage control, calibration, class-imbalance handling, interpretability, uncertainty reporting and deployment/actionability. Bibliometric and NLP analyses provided complementary field mapping. A second reviewer independently assessed 45 studies from the original 223-study audit subset across all ten audit domains. Results The literature expanded rapidly after 2021, with performance modelling and injury-risk prediction as the dominant applications. Studies frequently used GPS/GNSS, session-RPE/internal load, wellness/sleep, HR/HRV, neuromuscular testing, IMU/accelerometry and video-derived features. Random forests, logistic/linear models and gradient boosting were the most common model families. Deployment-relevant validation remained limited. After targeted reassessment and updated-search incorporation, temporal validity was strongly documented in 28 studies (10.9%), while independent external validation was strongly documented in one study (0.4%) and partially documented in one additional study (0.4%). Leakage control, calibration, repeated-observation independence and model-level uncertainty reporting were also uncommon. Pre-correction exact agreement in the independent audit subset was 65.3% overall and varied substantially across domains. Discussion Predictive modelling in team-sport athlete monitoring is expanding, but model development has advanced faster than trustworthy validation practice. Future studies should define the intended decision context, preserve temporal and player-level data structure, prevent leakage, report calibration and uncertainty, validate models under realistic deployment conditions and translate outputs into practitioner-facing decision support.
Authors: Šimon Salaj, Michal Hrubý, Jan Cacek
Institution: Masaryk University
Language: English
Subfield: Orthopedics and Sports Medicine
Field: Medicine
Domain: Health Sciences
Open Access status: gold
Estimated APC paid: $2,353
OpenAlex ID: W7225782241