mjboothaus/titanic-databooth

Dataset Description

Purpose: Demonstrate how data quality impacts analytics through the iconic Titanic dataset, featuring:

• —Original datasets (with known age/class errors)
• —Corrected versions (with reconciled passenger details)
• —Data quality annotations (error flags, reconciliation sources)

Homepage: Data Governance: Titanic Dataset and the Perils of Bad Data Repository: mjboothaus-titanic-databooth Tasks: data-cleaning, error-detection, survival-prediction

Dataset Versions

Data Fields (Corrected Version)

Usage Example

from datasets import load_dataset

# Compare original vs corrected data
original = load_dataset("mjboothaus/titanic-databooth", name="original")
corrected = load_dataset("mjboothaus/titanic-databooth", name="corrected-v1")

# Find corrected records
discrepancies = corrected.filter(lambda x: x["is_age_discrepancy"])
print(f"Fixed {len(discrepancies)} age errors")

<!-- TODO: Also st.connnector class? And "plain" class too for DuckDB? -->

Key Data Quality Issues

1. 1.Age Discrepancies
2. 2.Original error: 80yo survivor (actual age 47)
3. 3.143+ passengers with >2 year age differences
4. 4.Systemic bias from death age vs voyage age confusion

1. 1.Class Misassignments
2. 2.Documented cabin class errors
3. 3.Impacts fare/survival correlation analysis

Reconciliation Process

1. 1.Source Alignment: Cross-referenced with:
2. 2.Encyclopedia Titanica
3. 3.Titanic Facts Network
4. 4.Historical voyage manifests

1. 1.Validation Methods:
2. 2.Age distribution analysis
3. 3.Survival rate by age cohort
4. 4.Source conflict resolution protocols

Impact Analysis

Suggested Use Cases

• —Data Quality Workshops: Compare original/corrected versions
• —Governance Training: Demonstrate error propagation
• —ML Robustness Tests: Train models on both versions

Citation

@dataset{titanic-databooth,
  author = {Michael J. Booth},
  title = {Titanic Data Quality Benchmark},
  year = {2025},
  publisher = {Hugging Face},
  version = {1.0.0}
}

Acknowledgements

• —Encyclopedia Titanica for reference data

Key Features to Highlight:

• —Version Control: Clear lineage between original/corrected data
• —Error Documentation: Specific examples with historical context
• —Impact Metrics: Quantifiable differences between datasets
• —Educational Focus: Designed for data governance training

<!-- TODO: Include a Jupyter / Marimo / Streamlit notebook -->

Code demonstrating:

1. 1.Age distribution comparisons
2. 2.Survival rate analysis by data version
3. 3.Simple ML model performance differences

References:

Original "datacard" see https://huggingface.co/datasets/mjboothaus/titanic-databooth/resolve/main/titanic3info.txt

• —[1] https://www.databooth.com.au/posts/data-quality-titanic/
• —[2] https://mjboothaus.wordpress.com/2017/07/11/did-a-male-octogenarian-really-survive-the-sinking-of-the-rms-titanic-2/

license: apache-2.0 ---

Sponsored by: [DataBooth.com.au](https://www.databooth.com.au).