Author(s): JU, Y., SUH, S.H., Institution: YONSEI UNIVERSITY, Country: KOREA, SOUTH, Abstract-ID: 250

INTRODUCTION:
Anti-doping programs aim to protect fairness in competition and athlete health; however, conventional testing approaches are costly, invasive, and limited by narrow detection windows. The Athlete Biological Passport (ABP) provides a longitudinal monitoring framework but requires repeated biological sampling and substantial resources, limiting scalability. The Athlete Performance Passport (APP) has emerged as a complementary, non-invasive approach based on longitudinal performance data, yet empirical validation using advanced artificial intelligence (AI) methods remains limited, particularly in elite weightlifting characterized by complex performance trajectories. This study evaluated an AI-driven APP framework for detecting performance anomalies in elite male weightlifting.
METHODS:
Competition records were obtained from the official website of International Weightlifting Federation (IWF), comprising 28,831 entries from 1998 to 2022. Recurrent neural networks were used to model individual performance trajectories and predict expected total lifted weight at each athlete’s final competition. Performance residuals, defined as deviations between observed and predicted outcomes, were extracted and combined with demographic and performance variables to train an XGBoost classifier for estimating doping sanction status. Model interpretability was assessed using explainable artificial intelligence techniques based on Shapley additive explanations.
RESULTS:
The recurrent neural network demonstrated strong predictive accuracy, achieving an R² of 0.726 with stable performance across bootstrap resampling. The classification model showed good discriminative ability, with a receiver operating characteristic area under the curve of 0.831 on the internal test set and 0.738 in external validation. Explainable AI analysis identified total lifted weight, body weight, and performance residuals as the strongest contributors to sanction classification, indicating that longitudinal deviations provided additional information beyond absolute performance levels.
CONCLUSION:
An AI-driven APP framework can effectively identify performance anomalies consistent with doping-related enhancement in elite male weightlifting. Performance-based longitudinal monitoring represents a scalable and cost-effective complement to biological testing and has practical relevance for applied sports science by supporting targeted anti-doping strategies, enhancing athlete health protection, and promoting fairness in competition.