📌 A Machine Learning, Deep Learning, and NLP-based Approach

📄 Published in PLOS ONE (2025):
Hasan, M. J., Sultana, J., Ahmed, S., & Momen, S.
Early detection of occupational stress: Enhancing workplace safety with machine learning and large language models
PLOS ONE, 20(6), e0323265 (2025)

Workplace stress is a critical issue that impacts employee well-being, organizational efficiency, and workplace safety. This repository provides cutting-edge AI solutions for detecting, analyzing, and mitigating occupational stress using a combination of:

✅ Machine Learning (ML) for stress detection models.
✅ Deep Learning (DL) using advanced neural networks.
✅ Natural Language Processing (NLP) to extract insights from survey data.
✅ Explainable AI (XAI) for interpretability in workplace safety decisions.
✅ Synthetic Data Generation to validate model generalizability.

Why does this matter?

• Stressed workers are more prone to workplace accidents and lower productivity.
• Organizations spend billions annually on absenteeism and medical costs due to work-related stress.
• Traditional stress assessments (e.g., surveys, self-reports) are slow, subjective, and lack predictive power.

Our AI-driven approach automates stress detection, enhances workplace safety, and enables proactive decision-making for both organizations and policymakers.

📦 Occupational-Stress-Safety
├── Main_results_holdout_val.ipynb      # Main results reported in paper using holdout validation 
├── explainable_ai/                     # XAI-based analysis  
│   ├── XAI_Occupational_Stress.ipynb   
│
├── SOTA_analysis/                      # State-of-the-art comparison  
│   ├── SOTA_Paper_1.ipynb  
│   ├── SOTA_Paper_2.ipynb  
│   └── ...  
│
├── domain_analysis/                    # LLM-driven analysis  
│   ├── LLM_anova_+_rfecv.ipynb  
│
├── cross-validation/                    # Model validation  
│   ├── cross-validation.ipynb  
│
├── anova/                               # ANOVA-based feature selection  
│   ├── anova.ipynb  
│
├── rfecv/                               # Recursive feature elimination  
│   ├── RFECV.ipynb  
│
├── synthetic_data_generation/           # Synthetic dataset generation  
│   ├── Synthetic_data_generation.ipynb  
│   ├── Synthetic_data_comparison.ipynb  
│
├── eda/                                 # Exploratory Data Analysis  
│   ├── EDA_Occupational_Stress.ipynb  
│
├── ablations/                           # Impact of feature selection  
│   ├── Ablation_no_RFECV.ipynb  
│   ├── Ablation_no_anova.ipynb  
│   ├── Ablation_no_zero_var.ipynb  
│
├── dataset/                             # Raw dataset  
│   ├── DIB dataset and codebook.xlsx  
│
├── synthetic_dataset/                   # Generated datasets  
│   ├── synthetic_data_tvae.csv  
│   ├── synthetic_data_gaussian_copula.csv  
│   ├── synthetic_data_ctgan.csv  
│   ├── synthetic_data_copulaGan.csv  
│
├── LICENSE                              # MIT License  
└── README.md                            # This file  

1. Clone the Repository

   git clone https://github.com/junayed-hasan/occupational-stress-ml.git
   cd Occupational-Stress-Safety

2. Create a Virtual Environment

   python -m venv venv
   source venv/bin/activate  # Mac/Linux
   .\venv\Scripts\activate   # Windows

3. Install Dependencies

   pip install -r requirements.txt

4. Run Jupyter Notebooks

   jupyter notebook

• Run eda/EDA_Occupational_Stress.ipynb to explore dataset characteristics.

• anova/anova.ipynb for ANOVA-based feature selection.
• rfecv/RFECV.ipynb for recursive feature elimination.
• cross-validation/cross-validation.ipynb to validate models.

• Run synthetic_data_generation/Synthetic_data_generation.ipynb to generate synthetic datasets.

• explainable_ai/XAI_Occupational_Stress.ipynb provides model interpretability.
• domain_analysis/LLM_anova_+_rfecv.ipynb utilizes NLP models for deeper insights.

• 📌 Achieved 90.32% Accuracy in occupational stress classification.
• 📌 Top predictors of stress: Workload, Manager Support, Job Role Clarity.
• 📌 LLMs outperformed traditional NLP models in stress classification tasks.
• 📌 Synthetic data generation proved effective, achieving ~89% accuracy on unseen test scenarios.

For full details, check:

• Main_results_holdout_val.ipynb (final results)
• ablations/ (impact of feature selection methods)

This research has direct implications for:
✔ HR & Workforce Management – Predict stress & reduce workplace accidents.
✔ Occupational Health & Safety Teams – Implement AI-driven monitoring.
✔ Government & Policymakers – Set data-driven workplace safety regulations.
✔ AI & Data Science Practitioners – Develop real-world stress detection models.

If you use this repository in your research, please cite:

@article{hasan2025early,
  title={Early detection of occupational stress: Enhancing workplace safety with machine learning and large language models},
  author={Hasan, Mohammad Junayed and Sultana, Jannat and Ahmed, Silvia and Momen, Sifat},
  journal={PLoS One},
  volume={20},
  number={6},
  pages={e0323265},
  year={2025},
  publisher={Public Library of Science San Francisco, CA USA},
  doi={https://doi.org/10.1371/journal.pone.0323265}
}

This project is licensed under the MIT License, allowing free use, modification, and distribution.

Contributions are welcome! To contribute:

1. Fork the repository.
2. Create a new branch (feature-new-feature).
3. Commit changes and open a pull request.

📌 Collaborators: Mohammad Junayed Hasan and Jannat Sultana

📌 Supervisors: Dr. Sifat Momen (sifat.momen@northsouth.edu, https://scholar.google.com/citations?user=sGVZEaAAAAAJ), Ms. Silvia Ahmed (silvia.ahmed@northsouth.edu, https://scholar.google.com/citations?user=T5jK--YAAAAJ&hl=en&oi=ao)

📌 Emails: junayedhasan100@gmail.com, jannatsultana187@gmail.com

📌 LinkedIn: https://www.linkedin.com/in/mjhasan21/, https://www.linkedin.com/in/jannat-sultana/

We appreciate your interest in our research and welcome discussions, collaborations, and feedback! 🚀