WFH Burnout — Feature Importance Analysis

Author: Samuel Pauly

Identifies the key behavioral predictors of work-from-home burnout using three machine learning classifiers, derives statistically validated intervention thresholds, and tests simplified model performance.

---

Files

File	Description
wfh_burnout_dataset.csv	Dataset (required)
main.py	Runs the full pipeline in one script
01_EDA.ipynb	Exploratory data analysis
02_Models.ipynb	Model training and feature importance
03_Threshold_Analysis.ipynb	Threshold derivation and bootstrap robustness
04_Simplified_Interactions.ipynb	Simplified models and interaction effects
requirements.txt	Python dependencies

---

Setup

1. Install dependencies

pip install -r requirements.txt

2. Place the dataset in the same folder as the code files

wfh_burnout_dataset.csv

---

How to Run

Option A — Single script (recommended)

Runs the entire analysis from start to finish:

python main.py

Estimated runtime: 2–3 minutes (bootstrap and permutation importance are the slowest steps). Progress is printed to the console at each step.

Option B — Notebooks individually

Run in order inside Jupyter:

01_EDA.ipynb
02_Models.ipynb
03_Threshold_Analysis.ipynb
04_Simplified_Interactions.ipynb

Note: notebook 3 loads CSV files saved by notebook 2, so they must be run in order.

---

Outputs

Figures

• fig_01_class_distribution.png — burnout class bar chart and pie chart
• fig_04_boxplots_by_class.png — feature distributions by burnout class
• fig_08_threshold_violins.png — violin plots with intervention threshold lines
• fig_09_bootstrap_importance.png — bootstrap feature importance with 95% CI
• fig_10_roc_comparison.png — ROC curve, full vs simplified model

CSVs

• lr_importance.csv, rf_importance.csv, svm_importance.csv — feature importance scores
• thresholds.csv — derived intervention thresholds per feature
• bootstrap_importance.csv — bootstrap mean importance and confidence intervals
• simplified_model_results.csv — simplified model performance comparison