Customer Insights — Exploratory Data Analysis

A full exploratory data analysis (EDA) of a customer dataset: audit data quality, clean it, profile customer segments, quantify relationships, test a churn hypothesis, and tell the data story with clear visualizations. This is the day-to-day work of a data scientist — explore, find insight, communicate it.

⚠️ Uses synthetic data with deliberately embedded structure (so the insights are real and reproducible); works on any customer CSV via --csv.

---

🔍 What the analysis covers

1. Data-quality audit — per-column dtype, missing values, uniqueness, and flags. Catches a real issue: a few impossible ages (−1, 0, 250).
2. Cleaning — impossible ages → median imputation.
3. Segment profiling — spend, tenure, engagement, and churn by segment.
4. Correlation analysis — relationships among age, tenure, spend, engagement.
5. Hypothesis test — does higher engagement reduce churn? (Churn by quintile.)
6. Visualization — four presentation-ready charts saved to outputs/.

📊 Key findings (from the synthetic data)

Finding	Evidence
Segments are distinct	Premium avg spend $395 & churn 15% vs Budget $117 & churn 45%
Spend grows with tenure	corr(tenure, spend) = +0.40
Engagement curbs churn	corr(engagement, churn) = −0.26; lowest quintile churns 53% vs 17% for the highest

Recommendation: invest in early engagement to lift retention and lifetime value.

🖼️ Charts

⚙️ Install & run

pip install -r requirements.txt

python examples/run_eda.py                     # synthetic data + charts + narrative
python examples/run_eda.py --csv customers.csv # your own data

✅ Validation

python tests/test_analysis.py

Confirms the audit detects bad ages, cleaning fixes them, segments are distinct, and the embedded relationships (tenure→spend positive, engagement→churn negative) are recovered.

🗂️ Project layout

src/
  data_loader.py   # CSV loader + synthetic customer generator
  analysis.py      # quality audit, cleaning, segment & correlation analysis
  plots.py         # presentation-ready charts
examples/
  run_eda.py       # full EDA pipeline with a written narrative of findings
outputs/           # generated charts
tests/
  test_analysis.py

🔭 Roadmap

• Add a Jupyter notebook walkthrough with inline commentary
• Build a churn-prediction model on the cleaned features
• Interactive dashboard (Plotly / Streamlit)

---

Author: Dennis Chen (陳雲皓) · LinkedIn · GitHub