ARTIFICIAL INTELLIGENCE–DRIVEN SPORTS SCIENCE INNOVATION IN SWITZERLAND: A MACHINE LEARNING–BASED SMART FRAMEWORK FOR PERSONALIZED TRAINING, ATHLETIC PERFORMANCE OPTIMIZATION, AND INJURY RISK REDUCTION

Abstract

Artificial intelligence is reshaping sports science by enabling continuous athlete monitoring, personalized training prescription, performance forecasting, and early injury-risk detection. This study developed an artificial intelligence–driven smart framework for Swiss sports organizations by integrating machine learning, wearable-sensor data, physiological indicators, training-load records, and injury histories. A multisource dataset was constructed from 320 competitive athletes representing football, athletics, cycling, skiing, and indoor team sports across Swiss university and regional training centers. Data were collected over 24 weeks from GPS devices, heart-rate monitors, inertial measurement units, athlete wellness questionnaires, medical screening records, and performance tests. The final dataset contained 18,740 athlete-session observations and 42 variables, including age, sex, sport type, training duration, total distance, sprint count, acceleration load, heart-rate variability, resting heart rate, sleep duration, perceived fatigue, recovery score, previous injury, body mass index, jump performance, and session rating of perceived exertion. After data cleaning, normalization, imbalance correction, and recursive feature selection, random forest, support vector machine, XGBoost, and artificial neural network models were trained using stratified ten-fold cross-validation. XGBoost achieved the best injury-risk classification performance, with 92.6% accuracy, 0.91 precision, 0.89 recall, a 0.90 F1-score, and an area under the receiver operating characteristic curve of 0.95. The neural network produced the strongest performance-prediction results, achieving an R² of 0.86, a root mean square error of 0.12, and a mean absolute error of 0.08. Compared with conventional training plans, the proposed framework improved predicted performance by 14.8%, reduced excessive workload exposure by 21.3%, and lowered estimated injury risk by 27.6%. Feature-importance analysis identified acute-to-chronic workload ratio, previous injury, heart-rate variability, sleep duration, and acceleration load as the most influential predictors. The findings indicate that an AI-supported, athlete-centered framework can enhance training individualization, improve performance outcomes, and support preventive sports medicine in Switzerland. The framework also incorporates model transparency, secure data governance, and privacy-preserving decision support, providing a scalable foundation for coaches, clinicians, and sports institutions within both elite and developmental sporting environments.