Who survived the Titanic, and why — quantified. A complete, reproducible exploratory data analysis of the titanic5 dataset (1,309 passengers, 14 features) with 95% Wilson confidence intervals, odds ratios, chi-square tests, Cramer's V, and Cohen's d. Ships an interactive Chart.js dashboard, a stakeholder analyst report (DOCX + PDF), a classical EDA PDF, and a fully-tested Python package for EDA, statistics, and visualisation.
🔴 Live demo · aneekhait.github.io/titanic-data-analysis
📖 Read the writeup inline: REPORT.md · 📊 Open the dashboard: dashboard/index.html · 📄 Download PDF: analyst report
Author: Aneek Hait · License: MIT · Dataset: titanic5 (Encyclopedia Titanica / Vanderbilt Biostatistics)
Most Titanic projects stop at "women and children first" with a single bar chart. This one:
- Quantifies every claim with 95% confidence intervals, odds ratios, and effect sizes — not just raw percentages.
- Reconciles counts across every stage (raw → cleaned → engineered → reported) with an automated validator.
- Ships three audience-tuned deliverables from one engineered DataFrame — interactive dashboard, narrative analyst report, classical PDF.
- Is actually reusable: a clean
src/package with 49 pytest tests, ruff, pre-commit, GitHub Actions, and a Dockerfile. - Reads cold: every statistical finding has a "What this means in plain English" callout.
The dashboard renders the same data interactively with hover tooltips, error bars (Wilson 95% CIs), scroll-spy navigation, and dark + light themes. Open the live demo →
| Output | Audience | Path |
|---|---|---|
| 📖 Analyst report — rendered Markdown | Reads inline on GitHub | REPORT.md ← start here |
| Interactive dashboard (Chart.js, dark + light themes, scroll-spy nav) | Anyone exploring the data | dashboard/index.html |
| Analyst report — DOCX (editable) | Stakeholders, hiring managers, peer review | reports/Titanic_Survival_Analyst_Report.docx |
| Analyst report — PDF | Same content, fixed-layout for distribution | reports/Titanic_Survival_Analyst_Report.pdf |
| Classical EDA report (generated via fpdf2) | Legacy / chart-heavy view | dashboard/titanic_eda_report.pdf |
A clean src/ layout: data loading, cleaning, feature engineering, EDA, statistics, plotting — all unit-tested.
git clone https://github.com/AneekHait/titanic-data-analysis.git
cd titanic-data-analysis
make install # install dependencies
make download # fetch the titanic5 CSV (~1 MB)
make eda # run the full EDA pipeline + write 8 charts
make dashboard # build the interactive HTML dashboard
make report # build the analyst DOCX + PDF
make test # 49 pytest testsOutputs land in dashboard/, reports/, and outputs/figures/.
| Factor | Finding |
|---|---|
| Overall | 38.2% of 1,309 passengers survived (500 lived, 809 perished) |
| Sex | Women 72.8% vs men 19.1% — odds ratio ≈ 11.3× |
| Class | 1st 62.0%, 2nd 42.8%, 3rd 25.5% (Cramer's V ≈ 0.31) |
| Class × Sex | 1st-class women 96.5% vs 3rd-class men 15.2% — an 81-pp gap |
| Age | Children (≤16) survived at 49.0%; "children first" was real but small (OR ≈ 1.7×) |
| Fare | Strongest numeric predictor (r = 0.247), but mostly a proxy for class |
| Embarked | Cherbourg 56.6% — confounded with class composition |
| Lifeboat (mechanism) | 98.6% of recorded boat occupants survived vs 2.6% without a boat record |
Effect-size ranking (single comparable scale, |r| or Cramer's V):
Sex ▰▰▰▰▰▰▰▰▰▰▰▰▰ 0.527 (Large)
Pclass ▰▰▰▰▰▰▰ 0.313 (Medium)
Fare ▰▰▰▰▰ 0.247 (Small)
Embarked ▰▰▰▰ 0.204 (Small)
Parch ▰▰ 0.083 (Negligible)
Age ▰ -0.031 (Negligible)
SibSp ▰ -0.028 (Negligible)
titanic-data-analysis/
├── src/ # Python package (installable, tested)
│ ├── config.py # Paths and constants
│ ├── data/
│ │ ├── loader.py # Download & load CSV
│ │ └── processing.py # clean_data + engineer_features
│ ├── analysis/
│ │ ├── eda.py # Survival rates, missing values
│ │ ├── statistics.py # Chi², t-test, ANOVA, Cohen's d, Cramer's V
│ │ └── inference.py # Wilson CIs, odds ratios, joint tables
│ └── visualization/
│ └── plots.py # Reusable Matplotlib plot functions
├── scripts/
│ ├── download_data.py # Fetch titanic5.csv
│ └── run_eda.py # CLI EDA entry point (click)
├── dashboard/ # Interactive Chart.js dashboard
│ ├── generate.py # Builds index.html
│ ├── generate_pdf.py # Builds the classical EDA PDF
│ ├── index.html # ← published dashboard
│ └── titanic_eda_report.pdf # ← published classical PDF
├── reports/ # Author-written analyst report
│ ├── build_analyst_report.py
│ ├── validate_dataset.py # Cross-stage count reconciliation
│ ├── Titanic_Survival_Analyst_Report.docx
│ └── Titanic_Survival_Analyst_Report.pdf
├── notebooks/
│ └── 01_exploratory_analysis.ipynb
├── tests/ # 49 pytest tests across 5 files
├── docs/ # Methodology, data dictionary, dev guide
│ ├── METHODOLOGY.md
│ ├── DATA.md
│ └── DEVELOPMENT.md
├── data/raw/ # titanic5.csv (gitignored)
├── Dockerfile
├── Makefile
├── pyproject.toml
├── requirements.txt
├── README.md # ← you are here
├── CHANGELOG.md
├── CONTRIBUTING.md
├── ROADMAP.md # What's still aspirational
├── LICENSE # MIT
└── AGENTS.md # Notes for AI coding agents
| Doc | What's inside |
|---|---|
| docs/METHODOLOGY.md | Analytical approach, statistical machinery, why each test was chosen |
| docs/DATA.md | Data dictionary, source, cleaning rules, engineered features |
| docs/DEVELOPMENT.md | Local setup, running the suite, adding a feature |
| reports/README.md | Index of generated reports and how to rebuild them |
| dashboard/README.md | What's in the dashboard, how to regenerate, theme notes |
| AGENTS.md | Repo-specific notes for AI agents (Claude Code, Cursor, etc.) |
| docs/DISCOVERABILITY.md | One-time GitHub settings (topics, Pages, social preview) to make the repo findable |
| CONTRIBUTING.md | How to propose changes |
| CHANGELOG.md | Recent improvements |
| ROADMAP.md | Outstanding ideas (ML modelling, Streamlit app, MLflow, etc.) |
Core: pandas, numpy, scipy, matplotlib, seaborn, click Reports: fpdf2 (classical PDF), python-docx (analyst DOCX), Chart.js (interactive dashboard) Tooling: pytest, ruff, pre-commit, GitHub Actions, Docker
- Dataset is gitignored.
make downloadfetches it fromhbiostat.orgon demand. - All numbers in every output trace to the same engineered DataFrame. Run
python reports/validate_dataset.pyto verify counts reconcile across raw → cleaned → engineered. - CI runs the full pytest suite on every push (.github/workflows/ci.yml).
- Docker:
docker build -t titanic-eda . && docker run titanic-eda.
The classical PDF report (dashboard/titanic_eda_report.pdf) is the original output — generated programmatically by generate_pdf.py. It's chart-heavy and section-by-section.
The analyst report (reports/Titanic_Survival_Analyst_Report.docx) is what I'd hand to a stakeholder — written as a narrative, with an executive summary, a glossary, and a "What this means in plain English" callout under every statistical finding. The DOCX is editable; the PDF is fixed-layout. The dashboard is the same content as an interactive page with hover tooltips, CIs as error bars, scroll-spy nav, and dark/light themes.
If you only have time for one: open the dashboard.
If you reference this work in research, teaching, or another project, please cite it:
@software{hait_titanic_eda,
author = {Hait, Aneek},
title = {Titanic Survival Analysis: Interactive EDA and Statistical Inference},
year = {2026},
url = {https://github.com/AneekHait/titanic-data-analysis},
license = {MIT}
}A machine-readable CITATION.cff is included so GitHub renders a "Cite this repository" button.
- The titanic5 dataset is curated by Encyclopedia Titanica and hosted by Vanderbilt Biostatistics.
- Statistical methodology references: Wilson 1927 (score interval), Cohen 1988 (effect sizes), Cramer 1946 (Cramer's V).
Built by Aneek Hait.
Keywords (for search)
titanic dataset, titanic survival analysis, titanic eda, exploratory data analysis, kaggle titanic, data science portfolio project, statistical inference, odds ratio, chi-square test, Wilson confidence interval, Cramer's V, Cohen's d, pandas, numpy, scipy, matplotlib, seaborn, Chart.js, interactive dashboard, python data analysis, reproducible research, machine learning preparation, feature engineering, hypothesis testing.