دانشگاه قم مطالعات عملکردی در روانشناسی ورزشی 3060-8023 2 3 2025 07 25 Machine Learning-Driven Prediction and Personalized Intervention for Athlete Burnout: Integrating Biometric and Psychological Markers Machine Learning-Driven Prediction and Personalized Intervention for Athlete Burnout: Integrating Biometric and Psychological Markers 63 76 3627 10.22091/frs.2025.13436.1085 FA Fatma Hilal Yagin Department of Biostatistics, Faculty of Medicine, Malatya Turgut Ozal University, Malatya, Türkiye. 0000-0002-9848-7958 Sameer Badri Al-Mhann Department of Physiology, School of Medical Sciences, Universiti Sains Malaysia, Kubang Kerian 16150, Kelantan, Malaysia; Kelantan, Malaysia. 0000-0001-9063-0185 Journal Article 2025 02 28 <strong>Objective</strong>: Athlete burnout, which is a multidimensional syndrome and consists of emotional exhaustion, reduced accomplishment, and sport devaluation, is associated with detrimental outcomes to performance and well-being. The aim of this study was to design and validate a machine learning model by combing real-time physiological monitoring and psychological testing to predict and prevent burnout in elite athletes. <strong>Method:</strong> We recorded multimodal data for 120 national-level athletes (60 males, 60 females) from three sports over a 6-month period; we obtained heart rate variability (HRV), salivary cortisol, sleep measures, and standardized burnout scales. An ensemble model of XGBoost and LSTM architectures had the best predictive performance (AUC-ROC = 0.91), which was significantly better than that of the traditional logistic regression (AUC-ROC = 0.72, p < 0.001). <strong>Results</strong>: Distinguishing physiological predictors were HRV (β = -0.34, p < 0.001), cortisol awakening response attenuation (β = 0.29, p = 0.001) and deep sleep reduction (β = -0.27, p = 0.001), with the relationships being moderated by TrL (pinteraction < 0.05). In the three-month implementation trial, the system prospectively identified 68% of the burnout cases early on (median lead time = 18 days); it decreased the incidence by 37% relative to the controls (OR = 0.43, 95% CI [0.28, 0.66]). The model had strong temporal stability (AUC drift < 0.02/month), but there is potential for decreased generalizability to recreational athletes and technology-based restricting widespread application. <strong>Conclusions</strong>: Our results demonstrate the potential for machine learning-empowered combinatory continuous biometric monitoring and psychological screening to operationalize burnout as a preventive (rather than reactive) challenge in elite sports. The model offers coaches and medical teams’ actionable information to intervene in an individualized manner, but future research is needed to examine menstrual cycle effects and to design cost-effective interventions in youth sports systems.  <strong>Objective</strong>: Athlete burnout, which is a multidimensional syndrome and consists of emotional exhaustion, reduced accomplishment, and sport devaluation, is associated with detrimental outcomes to performance and well-being. The aim of this study was to design and validate a machine learning model by combing real-time physiological monitoring and psychological testing to predict and prevent burnout in elite athletes. <strong>Method:</strong> We recorded multimodal data for 120 national-level athletes (60 males, 60 females) from three sports over a 6-month period; we obtained heart rate variability (HRV), salivary cortisol, sleep measures, and standardized burnout scales. An ensemble model of XGBoost and LSTM architectures had the best predictive performance (AUC-ROC = 0.91), which was significantly better than that of the traditional logistic regression (AUC-ROC = 0.72, p < 0.001). <strong>Results</strong>: Distinguishing physiological predictors were HRV (β = -0.34, p < 0.001), cortisol awakening response attenuation (β = 0.29, p = 0.001) and deep sleep reduction (β = -0.27, p = 0.001), with the relationships being moderated by TrL (pinteraction < 0.05). In the three-month implementation trial, the system prospectively identified 68% of the burnout cases early on (median lead time = 18 days); it decreased the incidence by 37% relative to the controls (OR = 0.43, 95% CI [0.28, 0.66]). The model had strong temporal stability (AUC drift < 0.02/month), but there is potential for decreased generalizability to recreational athletes and technology-based restricting widespread application. <strong>Conclusions</strong>: Our results demonstrate the potential for machine learning-empowered combinatory continuous biometric monitoring and psychological screening to operationalize burnout as a preventive (rather than reactive) challenge in elite sports. The model offers coaches and medical teams’ actionable information to intervene in an individualized manner, but future research is needed to examine menstrual cycle effects and to design cost-effective interventions in youth sports systems.  https://frs.qom.ac.ir/article_3627_2e292a82fb9c4aa423510b1df341d688.pdf