KMAN Publication Inc. (KMANPUB) Journal of Adolescent and Youth Psychological Studies (JAYPS) 2981-2526 7 Serial Number 43 2026 03 10 Algorithmic Identification of Academic Burnout: Integrating Wearable Sensor Data and School Performance Metrics via Gradient Boosted Decision Trees Algorithmic Identification of Academic Burnout: Integrating Wearable Sensor Data and School Performance Metrics via Gradient Boosted Decision Trees 1 10 10.61838/kman.jayps.4174 EN Journal Article 2025 11 19 <table style="height: 815px; width: 104.04%;"> <tbody> <tr> <td style="width: 99.2233%;"> <p><strong>Objective:</strong> To develop and validate a multimodal machine learning pipeline utilizing Gradient Boosted Decision Trees to accurately identify academic burnout by integrating continuous physiological data from wearable sensors with behavioral school performance metrics.</p> <p><strong>Methods and Materials:</strong> A prospective longitudinal study was conducted with a cohort of undergraduate students enrolled in high-intensity programs in Brazil. Multimodal data collection encompassed continuous physiological monitoring via wrist-worn sensors to capture heart rate variability, electrodermal activity, and sleep architecture, alongside institutional academic registry data tracking lecture attendance, assignment submission latency, and examination scores. Ground-truth burnout classification was established using the Maslach Burnout Inventory-Student Survey. A Gradient Boosted Decision Trees (GBDT) algorithm was trained and optimized to classify students into burnout and non-burnout categories, evaluating performance through accuracy, F1 score, and area under the receiver operating characteristic curve (AUC-ROC).</p> <p><strong>Findings:</strong> The demographic analysis revealed a burnout prevalence of ( ) within the sample. The optimized GBDT model demonstrated superior predictive capability, achieving an overall classification accuracy of , an F1 score of , and an AUC-ROC of . Feature importance analysis indicated that physiological indicators of allostatic load, primarily rolling heart rate variability (  relative importance), and behavioral withdrawal metrics, specifically assignment submission latency (  relative importance), were the most significant predictors of the syndrome. Furthermore, the burnout cohort exhibited marked autonomic dysregulation, characterized by severely reduced deep sleep and elevated electrodermal activity peaks, directly mirroring their concurrent academic decline.</p> <p><strong>Conclusion:</strong> The integration of continuous biometric sensor data with institutional academic metrics via gradient boosting algorithms provides a highly accurate, objective framework for the early identification of academic burnout.</p> </td> </tr> </tbody> </table> https://journals.kmanpub.com/index.php/jayps/article/download/4174/9342