Abstract
Occupational burnout and chronic psychological distress have become critical systemic challenges within academic and corporate ecosystems. Conventional diagnostic frameworks primarily utilize psychometric instruments, such as the Maslach Burnout Inventory (MBI); however, these self-reporting methodologies are often compromised by subjectivity, social desirability bias, and retrospective recall inaccuracies. This thesis presents a novel computational framework that utilizes Facial Emotion Recognition (FER) to provide an objective, longitudinal, and non-intrusive quantification of burnout markers.
The architecture employs a hybrid deep learning approach, integrating Convolutional Neural Networks (CNNs) for high-dimensional spatial feature extraction with Long Short-Term Memory (LSTM) networks to capture temporal dependencies in facial micro-expressions. The pipeline ingests real-time video streams, maps facial configurations to discrete emotion probability distributions, and derives a composite stress index by applying weighted coefficients to negative valence states.
To ensure empirical rigor, the system’s predictive outputs are benchmarked against validated psychometric data. A centralized analytics dashboard facilitates the visualization of cross-sectional and multi-session stress trajectories. Pearson correlation analysis reveals a significant statistical convergence between FER-derived metrics and established burnout scales. This research underscores the efficacy of biometrically-driven affective computing as a scalable diagnostic tool for psychological health monitoring in high-stakes environments, such as healthcare and pedagogy.
Keywords
- Facial Emotion Recognition (FER)
- Burnout Detection, Prolonged Stress
- Convolutional Neural Network (CNN)
- Long Short-Term Memory (LSTM)
- Deep Learning,
- Machine Learning
- Computer Vision
- Transformers
- Histogram of Oriented Gradients (HOG)
- Support Vector Machine (SVM)
- Principal Component Analysis (PCA)
- Sequence Model
- Pipeline
- Pooling Layers.
Notes
Due to the confidentiality and security issues, the undergraduate thesis is not publicly available at the moment. Therefore, please contact me through the contact section or email to request for the paper for either personal use or research puposes. However, the presentation slide is attached at this page and is readily available for download.
Credits
- Institution: Seoul National University
- Thesis Advisor: Prof. Chang-Gun Lee