Globe Bank International — Customer Retention Analysis

Python Portfolio Project | End-to-End Customer Churn Analysis

🎯 The Challenge

Banks face a costly and persistent problem: customers leave without warning. As a retention manager, you might only see monthly reports that arrive too late. Key questions include:

• Which customers are most likely to churn next month?
• Are certain demographic groups leaving at higher rates?
• What behaviours distinguish churned customers from retained ones?
• Where should the retention budget be prioritised for maximum impact?

This was the business problem I set out to solve.

💡 My Solution

I built a four-stage Python analysis pipeline that transforms raw customer data into actionable retention insights. The solution moves from data exploration to polished dashboards and prioritised recommendations — demonstrating a full analytical workflow.

📊 Project Architecture

Stage 1 — Setup and Exploration

Notebook: 01_setup_exploration.ipynb

• Loaded 10,127 customer records across 23 raw columns
• Dropped 2 Naive Bayes classifier columns and the CLIENTNUM identifier
• Confirmed zero missing values and duplicates
• Encoded Attrition_Flag (0 = Retained, 1 = Churned)
• Visualised churn distribution, numeric histograms, and categorical breakdowns

Key finding: 16.1% churn rate — 1,627 customers left out of 10,127.

Stage 2 — Cleaning and Transformation

Notebook: 02_cleaning_transformation.ipynb

Encoding applied:

Features engineered:

Output: data/processed/bank_churners_clean.csv — 10,127 rows × 30 columns

Stage 3 — Analysis and Insights

Notebook: 03_analysis_insights.ipynb

Techniques used:

1. Correlation Analysis — Total_Trans_Ct, Total_Trans_Amt, and Trans_Freq_Ratio predict retention; Months_Inactive_12_mon and Contacts_Count_12_mon predict churn
2. Churn by Demographics — Platinum card holders and lower-income segments churn more; single customers slightly more than married
3. Statistical Comparisons — Welch’s t-tests with Cohen’s d for numeric features
4. Random Forest Feature Importance

Top 5 churn predictors:

Insight: Transaction behaviour dominates churn prediction over demographics.

Stage 4 — Visualisation and Storytelling

Notebook: 04_visualisation_storytelling.ipynb

Outputs:

🗂️ Project Structure

globe-bank-retention-analysis/
├── data/
│   ├── raw/
│   └── processed/
├── notebooks/
│   ├── 01_setup_exploration.ipynb
│   ├── 02_cleaning_transformation.ipynb
│   ├── 03_analysis_insights.ipynb
│   └── 04_visualisation_storytelling.ipynb
├── outputs/
│   ├── figures/
│   └── reports/
├── .gitignore
├── README.md
└── requirements.txt

🛠️ Tools and Libraries

💡 Business Recommendations

🎓 Reflections and Key Takeaways

Technical lessons:

• Transaction behaviour dominates demographic features
• Effect size matters as much as p-values
• Feature engineering captures real behavioural patterns

Design lessons:

• Storytelling is separate from analysis
• Consistent visual design builds trust

Process lessons:

• Working notebook by notebook keeps complexity manageable
• A beginner can produce portfolio-ready, structured, and clean work

📂 Project Resources

• GitHub Repository: View Full Code and Notebooks
• Includes Jupyter notebooks, cleaned dataset, charts, and interactive HTML reports

📧 Let’s Connect

• Portfolio: https://moshraf.github.io/
• GitHub: https://github.com/moshraf

• 2026 2
• 2025 1
• 2023 5