Early detection of student burnout using data science: a study of behavioral and psychological indicators

While burnout is being increasingly noted in educational contexts, the existing research primarily focuses on professional settings. Rising academic demands have exposed students to high workloads and psychological distress. Academic burnout is defined as exhaustion, disengagement, and low feelings of professional competence, which negatively affect students' well-being, motivation, and subsequent career prospects. In this study, we aimed to examine lifestyle and psychological predictors to help explain burnout among high school and university students in Pakistan, as well as the potential application of machine learning in early detection. We hypothesized that sleep, screen time, and mental exhaustion would each be robust predictors of the risk of burnout, and that predictive modeling has the capacity to classify at-risk students accurately. We used an online survey with 74 high school and university students (ages 12-25) to measure demographics, sleep duration, screen exposure, coping strategies, motivation, and fatigue. We analyzed the data using statistical analyses, and applied the scikit-learn machine learning module in Python for predictive modeling. Overall, sleep duration was negatively related to burnout, while mental fatigue was the strongest positive predictor of burnout. Screen time exposure only had a weak, non-significant association with burnout once other variables were accounted for. Additionally, there were some gender differences, with males exhibiting lower levels of burnout, although effect sizes were modest. Lastly, machine learning predictive models were accurate at classifying risk for students, with particular success from the Random Forest algorithm, demonstrating the potential for predictive analytics in the well-being of students.