BrotherTony/employee-burnout-turnover-prediction-800k

---

Synthetic Employee Dataset

800,000+ employee records with real-world distributions for burnout prediction, turnover analysis, and HR analytics

---

What You Get

This isn't just another CSV dump. You're looking at 800K+ carefully engineered employee profiles that mirror actual workforce dynamics, complete with performance metrics, burnout indicators, skill matrices, and behavioral personas. Think of it as a production-ready HR database that never existed but feels like it should have.

The sweet spot? All the statistical realism of real employee data, zero privacy concerns, zero compliance headaches.

---

Quick Stats

Total Records	800,000+
Departments	38 unique divisions
Job Roles	300+ distinct positions
Employee Personas	12 behavioral archetypes
Features per Record	30+ attributes
Salary Range	$27K - $384K

---

Why This Dataset is really really great

For Data Scientists

• Turnover prediction models ready to train out of the box
• Multi-class classification on employee personas
• Regression tasks for salary, performance, and burnout scoring
• Time-series potential with tenure and progression metrics

For HR Analytics

• Benchmark your workforce against 800K profiles
• Identify burnout patterns before they become exits
• Map skill gaps across departments and roles
• Test retention strategies without risking real employee data

For Researchers

• Study work-life balance dynamics at scale
• Analyze communication patterns and team sentiment
• Explore correlations between workload, satisfaction, and performance
• Validate your models on realistic synthetic populations

🌟 Community Use Cases

1. DSE 220: Employee Burnout Analysis By Grant Wagener, Duy Nguyen, and Thomas Brehme (UCSD) This is a great example of an end to end ML project using this dataset. The team did a full analysis, fixed some class imbalances, and built a segmented Random Forest model.

They actually managed to get 91% recall on identifying preventable turnover (burnout/salary issues). Worth a look if you want to see how to handle the data cleaning and feature engineering steps.

https://github.com/Harry-Wagener/DSE-220

2. iMML Library Tutorial The official tutorial for this Python library uses this dataset to teach Multimodal Learning.

https://github.com/ocbe-uio/imml

---

Dataset Anatomy

Click to expand full schema (30+ fields)

Core Identity

• employee_id Unique synthetic identifier
• role Job title (300+ varieties)
• job_level Entry | Mid | Senior | Manager | Lead
• department Business unit (38 departments)
• tenure_months Time with company (1-357 months)

Compensation

• salary Annual USD ($27K-$384K range)

Performance Indicators

• performance_score Normalized 0-1 rating
• project_completion_rate Delivery success metric
• goal_achievement_rate Objective completion
• role_complexity_score Job difficulty rating
• career_progression_score Growth trajectory

Wellbeing & Engagement

• satisfaction_score Employee happiness (0-1)
• workload_score Workload burden indicator
• stress_level Stress assessment (0-1)
• burnout_risk Burnout probability (0-1)
• training_participation L&D engagement

Collaboration & Communication

• team_sentiment Team morale (0-1)
• email_sentiment Email tone analysis
• slack_activity Chat platform usage
• meeting_participation Meeting engagement
• collaboration_score Teamwork rating
• communication_patterns Persona-based style

Skills & Competencies

• technical_skills Array of tech skills (4-9 per employee)
• soft_skills Array of soft skills (3-7 per employee)

Attrition & Risk

• left_company Boolean exit flag
• turnover_reason Exit category (if applicable)
• turnover_probability_generated ML-predicted churn risk
• risk_factors_summary Low | Medium | High | Severe risk

Behavioral Profiling

• persona_name 12 archetypes:
  • OverachievingSprinter
  • BurntOutStar
  • QuietAchiever
  • NewEnthusiast
  • StrugglingLearner
  • SteadyEddy
  • ChangeResistor
  • SeasonedExpert
  • SocialCatalyst
  • 9-to-5Clockwatcher
  • AmbitiousClimber
  • CorporateNavigator (Persona Names are A.I. generated)

Qualitative Data

• recent_feedback Anonymized employee comments (1000+ unique snippets)
• overtime_hours Extra hours logged (0-74 range)

---

Sample Use Cases

# imports
from datasets import load_dataset
import pandas as pd
import seaborn as sns
import matplotlib.pyplot as plt
from sklearn.linear_model import LogisticRegression

# load the dataset
dataset = load_dataset("BrotherTony/employee-burnout-turnover-prediction-800k")
df = dataset["train"].to_pandas()

# predict who's about to quit (simple logistic regression)
X = df[['satisfaction_score', 'burnout_risk', 'workload_score']]
y = df['left_company']

model = LogisticRegression()
model.fit(X, y)

# find your burnout hotspots
burnout_hotspots = df.groupby('department')['burnout_risk'].mean().sort_values(ascending=False)
print(burnout_hotspots.head())

# identify skill gaps by role
skill_gaps = df.explode('technical_skills').groupby(['role', 'technical_skills']).size()
print(skill_gaps.head())

# correlation deep dive
sns.heatmap(df[['performance_score', 'satisfaction_score', 
                'workload_score', 'burnout_risk']].corr(),
            annot=True, cmap="coolwarm")
plt.show()

---

Technical Details

Format: Parquet (optimized for analytics)
Encoding: UTF-8
Generation Method: Synthetic-real hybrid using distribution matching and constraint satisfaction

---

Licensing & Usage

License	CC-BY-NC-ND 4.0
Allowed	Research • Education • Academic papers • Internal analysis • Model training • Portfolio projects
Not Allowed	Commercial products • Selling the dataset • Redistributing modified versions
Attribution	Required (cite this dataset)

TL;DR: Free for learning, research, and experimentation. Just give credit and don't sell it.

---

Pro Tips

1. Start with personas The 12 behavioral archetypes are your cheat code for segmentation
2. Burnout + satisfaction isn't linear High performers can have high burnout (see: OverachievingSprinter)
3. Feedback text is gold 1000+ unique anonymized comments perfect for sentiment analysis
4. Skills are arrays Don't forget to explode those columns for proper analysis
5. Turnover probability ≠ actual turnover Two different signals, both useful

---

FAQ

Q: Is this real employee data?
A: No. Zero real people. Synthetic generation based on real-world distributions and patterns.

Q: Can I use this for my startup's HR tool?
A: Not commercially. Research and internal testing only.

Q: Why are some salaries weird?
A: Reflects real market variance. Geography, experience, and role complexity create wide ranges.

Q: What's with the personas?
A: Behavioral archetypes based on performance + satisfaction + workload combos. Useful for clustering and prediction.

Q: Can I redistribute this?
A: No derivatives or redistribution. Link back here instead.

---

Citation

If you use this dataset in your research or project, please cite:

@dataset{employee-burnout-turnover-prediction800k,
  author = {BrotherTony},
  title = {Synthetic Employee Dataset: 800K+ Records for HR Analytics},
  year = {2025},
  publisher = {Hugging Face},
  url = {https://huggingface.co/datasets/BrotherTony/synthetic-employee-dataset}
}

---

Why does this dataset exist?

The Story Behind This Dataset

This wasn't built in a weekend sprint or thrown together for clout. This dataset exists because of a very specific mission that evolved into something bigger.

The Original Goal

Started this whole thing to build something commercially viable - specifically an AI product called the "Prevue Explainer." The stakes? Nothing less than securing a comfortable retirement for my parents, supporting my brothers, and eventually stepping away from the commercial world to live as a full-time pilgrim dedicated to service and learning. High stakes meant high standards. Every data point, every structural decision, every edge case - all of it got the obsessive attention you'd give something when your family's future depends on it. That's why the psychological depth and strategic architecture here goes deeper than your typical synthetic dataset.

The Pivot

Somewhere in the trenches - debugging impossible errors at 3 AM, architecting logic systems, wrestling with data integrity - something clicked. The process itself became the point. Building this thing awakened a genuine passion for the craft that went beyond the original mission. Then came the bigger realization: the most valuable thing I'd created wasn't the potential product. It was the knowledge base itself. The reasoning frameworks, the ethical guardrails, the expert-level structures - all of it could do more good as a public resource than locked behind a commercial wall.

Why Open Source??

So here we are. The mission didn't die - it evolved. Instead of one proprietary tool helping a limited audience, this dataset (and others like it) can help researchers, students, data scientists, and organisations anywhere. More direct impact. Wider reach. Better outcomes. This is one of several datasets from that original project. Might be an earlier version since I'm still hunting down the others, but the core quality and intentionality? That's all here. If you use this and it helps your work, that's the whole point. That's the mission fulfilled.

Feedback & Contact

Spot something weird? Have ideas for v2? Open an issue or reach out through HuggingFace.

Built for the data science and HR analytics community.

---

Citation

If you use this dataset, please cite: BrotherTony (2025). Employee Burnout & Turnover Prediction 800k [Dataset]. Hugging Face. License: CC-BY-NC-ND 4.0

Made with <3 | Employee harm Free™