| Predicting Athletes' Physiological Performance Using Artificial Intelligence Models |
| Paper ID : 1630-SPORTCONGRESS |
| Authors |
|
Najmeh Barghi * Ministry of Education |
| Abstract |
| The integration of artificial intelligence (AI) into sports science has transformed how researchers and practitioners analyze, predict, and optimize athletes' physiological performance. This review synthesizes studies applying machine learning (ML), deep learning (DL), and hybrid AI systems to predict key physiological parameters such as maximal oxygen uptake (VO₂max), lactate threshold, fatigue onset, recovery kinetics, and injury risk in athletic populations. It draws on over two decades of research and emphasizes modeling techniques (neural networks, support vector machines, ensemble methods) and emerging DL architectures (e.g, convolutional and recurrent neural networks) increasingly applied in sports performance contexts. Recent reviews highlight that DL approaches have been especially successful for movement classification, tracking, and performance prediction across both individual and team sports. Results show that AI models often achieve high predictive accuracy under controlled laboratory or sport specific conditions (e.g, integrating sensor and video data), yet their generalization into real world field settings and across athlete populations remains limited. Data challenges (small sample sizes, heterogeneous measurements, subgroup imbalance), ethical concerns (data privacy, interpretability of black-box models), and limited transparency are recurrent. For instance, a systematic review reported moderate-to-large effect sizes for AI-based performance interventions (SMD ≈ 0.68) but also moderate heterogeneity (I² ≈ 58%). Key gaps include the need for multimodal data fusion (physiological, biomechanical, contextual), explainable AI frameworks to enhance coach and athlete trust, and cross-domain, longitudinal validation to improve model robustness. Recent works emphasize the scalability of DL across different sports and stress the necessity of rigorous methodology and standardized outcomes. In conclusion, while AI holds considerable promise for personalized athlete monitoring and performance optimization, future research should prioritize real world deployment, interpretable modeling, and inclusive datasets to fully harness its potential in high performance and exercise science. |
| Keywords |
| Artificial Intelligence, Physiological Performance, Sports Analytics, Machine Learning, Athlete Monitoring |
| Status: Abstract Accepted (Poster Presentation) |
