Early Detection of Academic Burnout Using Student Interaction Logs and Machine Learning

Abstract

Academic burnout is a growing concern in today\'s digitally driven education system. The prolonged stress of online learning, frequent assessments, and lack of physical interaction has increased the risk of mental fatigue among students. This paper presents a machine learning–based approach to detect early signs of academic burnout using student interaction logs collected from online platforms. The proposed model analyses behavioural data such as login frequency, study time, quiz performance, and emotional feedback to predict burnout risk levels. Early detection allows timely intervention from educators and counsellors, improving student well-being and academic performance.

Introduction

Recent shifts to online and hybrid education—accelerated by the COVID-19 pandemic—have increased student flexibility but also introduced mental health challenges, especially academic burnout. Burnout includes emotional exhaustion, low motivation, and declining performance, which can lead to anxiety, poor grades, or even dropout.

Traditional methods like teacher observation or surveys are often slow and ineffective, especially in large or remote classes. To address this, the paper proposes a smart system that uses machine learning (ML) and student interaction logs to detect early signs of burnout automatically and non-intrusively.

Key Components of the Proposed System

1. Data Sources

Collected from a simulated online platform:

• Login/logout times

• Time spent on lectures/quizzes

• Assignment and quiz scores

• Self-reported stress levels (optional)

• Facial expressions via webcam

2. Feature Engineering

Key indicators used for detecting burnout:

• Irregular study times

• Decline in performance

• Frequent late submissions

• Emotional signs (sadness, frustration)

• Session gaps and inconsistent activity

3. Machine Learning Models Used

• Logistic Regression – simple and interpretable

• Random Forest – ensemble-based, accurate

• Support Vector Machine (SVM) – good for complex data

• XGBoost – high performance, often best results

4. Evaluation Metrics

• Accuracy – overall correctness

• Precision & Recall – detection quality

• F1-Score – balanced measure

• ROC-AUC – model's ability to distinguish risk levels

System Architecture

1. Log Collection – student activity data from learning platform

2. Feature Extraction – behavioural and emotional indicators

3. ML Prediction – classifies burnout as Low, Moderate, or High

4. Burnout Alert – sends notifications to mentors or students

5. Intervention – suggests breaks, check-ins, or support resources

Key Results

• Patterns such as low productivity, frequent task-switching, and emotional fatigue (from facial cues) effectively predicted burnout.

• The system enabled early intervention, improving students’ academic performance and emotional well-being.

• Best performance was often achieved with Random Forest and XGBoost.

Real-World Applications

• University Learning Platforms (e.g., Moodle) – integrated burnout tracking

• Counseling Services – automated student alerts

• Parental Feedback Systems – monthly wellness reports

• Student Apps – reminders for breaks and self-care

Conclusion

This study shows that machine learning can be a powerful tool to detect early signs of academic burnout by analyzing students\' behaviour on online platforms like how often they log in, how long they study, or how their performance changes over time. By noticing these early warning signs, the system can alert teachers or mentors before the student’s condition gets worse. Early detection means students can get help such as counselling, study support, or simple motivation at the right time. This not only helps students feel better mentally and emotionally but also improves their academic performance and confidence. Such systems are especially useful in today’s digital learning environment, where teachers may not always notice when a student is struggling. By using technology in a smart and thoughtful way, we can create a learning space that doesn’t just focus on marks but also takes care of students\' mental well-being. In the future, combining this system with real-time emotion tracking, wearable devices, or mobile apps could make it even more effective and accessible to students everywhere.

References

[1] A. P. Singh and S. A. Sawant, “Smart Learning with Stress Detection using AQG and FER,” IEEE Xplore, 2024. [DOI: 10.1109/11089472]. [2] N. Gupta et al., “Predicting Academic Fatigue through Sensor and Log Data,” IEEE Conference on ML in Education, 2023. [DOI: 10.1109/10433060]. [3] S. Sharma et al., “Hybrid Model for Early Detection of Student Burnout,” IEEE Trans. Learning Analytics, 2024. [DOI: 10.1109/10825089]. [4] R. K. Sinha et al., “Facial Emotion Analysis for Burnout Prediction,” IEEE AI in Education, 2024. [DOI: 10.1109/11031490]. [5] A. M. Injadat, A. Moubayed, A. B. Nassif, and H. Lutfiyya, “Student Engagement Level in an e?Learning Environment: Clustering Using K?means,” in *Proc. IEEE World Engineering Education Conference (EDUNINE)*, Buenos Aires, Argentina, 2018, pp. 1–6. [6] A. Moubayed, M. Injadat, A. Shami, and H. Lutfiyya, “Relationship Between Student Engagement and Performance in E-Learning Environment Using Association Rules,” in *Proc. IEEE World Engineering Education Conference (EDUNINE)*, Buenos Aires, Argentina, 2018, pp. 1–6. [7] N. Gao, W. Shao, M. S. Rahaman, and F. D. Salim, “n?Gage: Predicting In?Class Emotional, Behavioural and Cognitive Engagement in the Wild,” *arXiv preprint*, Jul. 2020. [Online]. Available: arXiv:2007.04831 [Accessed: Jun. 2025]. [8] P. Sharma, S. Joshi, S. Gautam, S. Maharjan, S. R. Khanal, M. C. Reis, and J. B. Filipe, “Student Engagement Detection Using Emotion Analysis, Eye Tracking and Head Movement with Machine Learning,” *arXiv preprint*, Sep. 2019. [Online]. Available: arXiv:1909.12913 [Accessed: Jun. 2025]. [9] [5] M. Geng, M. Xu, Z. Wei, and X. Zhou, “Learning Deep Spatiotemporal Feature for Engagement Recognition of Online Courses,” in *Proc. 2019 IEEE Symposium Series on Computational Intelligence (SSCI)*, Xiamen, China, 2019, pp. 442–447. [10] A. V. Alvarez Jr., “Investigating the Association Between Student Engagement With Video Content and Their Learnings,” *IEEE Trans. Educ.*, 2023. [Online]. Available: doi:10.1109/TE.2023.3303421 [Accessed: Jun. 2025]. [11] [7] Abdallah Moubayed, Andrew Injadat, Abdallah Shami, Ali B. Nassif, and Hanan Lutfiyya, “Multi?split optimized bagging ensemble model selection for multiclass educational data mining,” *Applied Intelligence*, vol. 50, pp. 4506–4528, Jul. 2020. [12] L. Hallauer, E. M. Davenport, and R. S. Sutton, “Predicting Student Engagement Using Sequential Ensemble Model,” *IEEE Trans. Learn. Technol.*, 2023, vol. ?, no. ?, pp. ?, DOI: 10.1109/TLT.2023.3342860. [13] O. M. Nezami, M. Dras, L. Hamey, and D. Richards, “Automatic Recognition of Student Engagement using Deep Learning and Facial Expression,” *arXiv preprint*, Aug. 2018. [Online]. Available: arXiv:1808.02324 [Accessed: Jun. 2025].

Copyright

Copyright © 2025 Ritik Bhat. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.