Detecting Burnout Among Undergraduate Computing Students with Supervised Machine Learning

Background: Academic burnout significantly impacts students’ cognitive and psychological well-being and may result in adverse behavioral changes. An effective and timely detection of burnout in the student population is crucial as it enables educational institutions to mobilize necessary support systems and implement intervention strategies. However, current survey-based detection methods are susceptible to response biases and administrative overhead. This study investigated the feasibility of detecting academic burnout symptoms using machine learning trained exclusively on university records, eliminating reliance on psychological surveys. Methods: We developed models to detect three burnout dimensions—exhaustion, cynicism, and low professional efficacy. Five machine learning algorithms (i.e., logistic regression, support vector machine, naive Bayes, decision tree, and extreme gradient boosting) were trained using features engineered from administrative data. Results: Results demonstrated considerable variability across burnout dimensions. Models achieved the highest performance for exhaustion detection, with logistic regression obtaining an F1 score of 68.4%. Cynicism detection showed moderate performance, while professional efficacy detection has the lowest performance. Conclusions: Our findings showed that automated detection using passively collected university records is feasible for identifying signs of exhaustion and cynicism. The modest performance highlights the challenges of capturing psychological constructs through administrative data alone, providing a foundation for future research in unobtrusive student burnout detection.