Bank Customer Churn Prediction App

Table of Contents

1. Overview
2. Project Highlights
3. Data Analysis and Insights
4. Features
5. Project Structure
6. Getting Started
   • Prerequisites
   • Installation
   • Docker Deployment
7. Usage
8. Technical Components
9. Model Performance
10. Example Scenario
11. Contributing
12. License

Overview

The Bank Customer Churn Prediction App is a comprehensive data science project designed to predict customer account closures (churn) using advanced machine learning techniques. This end-to-end solution transforms raw customer data into actionable insights, helping financial institutions proactively manage customer retention.

Project Highlights

• 🤖 Advanced ML Ensemble: XGBoost, LightGBM, and CatBoost
• 📊 High Prediction Accuracy: AUC Score of 0.891
• 🌐 Web Interface: Flask-based predictive application
• 🐳 Containerized Deployment: Docker support
• 📈 Comprehensive Data Analysis: In-depth exploratory analysis

Technologies Used

• Language: Python 3.11.5
• Core Libraries:
  • Pandas, NumPy
  • Scikit-learn
  • Matplotlib, Seaborn

Web & Deployment

• Framework: Flask
• Containerization: Docker

Demonstration

Data Analysis and Insights

Key Findings

• Churn Rate: 21% of customers in the dataset
• Geographic Trends:
  • Highest churn rate in Germany
• Demographic Observations:
  • Females showed higher churn propensity
• Customer Activity:
  • Non-active members more likely to churn

Feature Engineering Highlights

• Created Age-Tenure Ratio
• Developed Active Credit Card Status
• Implemented targeted encoding strategies

Visualization Insights

1. Zero-balance accounts have higher churn probability
2. Credit card ownership minimally impacts churn
3. Customers with 3-4 products exhibit increased churn risk

Features

Machine Learning

• Ensemble Classifiers:
  • XGBoost
  • LightGBM
  • CatBoost
• Advanced preprocessing techniques
• Hyperparameter optimization

Web Application

• Intuitive user interface
• Real-time churn prediction
• Sample data auto-fill functionality

Deployment

• Docker containerization
• Easy-to-use deployment

Project Structure

.
├── README.md                   # Project documentation
├── app/                        # Flask application
│   ├── app.py                  # Main application entry point
│   ├── static/                 # Static web resources
│   │   └── styles.css          # Application styling
│   ├── templates/              # HTML templates
│   │   ├── base.html
│   │   ├── form.html
│   │   └── result.html
│   └── utils.py                # Utility functions
├── models/                     # Machine learning models
│   ├── train.py                # Model training script
│   ├── test.py                 # Model testing script
│   └── model.joblib            # Trained model
├── requirements.txt            # Python dependencies
└── dockerfile                  # Docker configuration

Getting Started

Prerequisites

• Python 3.10+
• Docker (optional)
• pip package manager

Installation

1. Clone the repository:

   git clone https://github.com/your-username/bank-churn-prediction.git
   cd bank-churn-prediction

2. Create virtual environment:

   python -m venv .venv
   source .venv/bin/activate  # Unix/macOS
   # or
   .venv\Scripts\activate     # Windows

3. Install dependencies:

   pip install -r requirements.txt

Docker Deployment

# Build Docker image
docker build -t churn-prediction-app .

# Run container
docker run -p 5000:5000 churn-prediction-app

# OR pull pre-built image
docker pull sayedgamal/churn-prediction-app:latest
docker run -p 5000:5000 sayedgamal/churn-prediction-app:latest

Usage

Web Application

1. Navigate to http://localhost:5000
2. Input customer details
3. Click "Predict" for churn probability

Jupyter Notebook

Explore detailed analysis in bank-customer-churn-prediction-0-891-auc-score.ipynb

Technical Components

• Preprocessing:
  • Target encoding
  • Feature scaling
  • Custom engineering pipelines
• Model Training:
  • Ensemble voting mechanism
  • Cross-validation
• Deployment:
  • Flask web framework
  • Docker containerization

Model Performance

• AUC Score: 0.891
• Algorithms: XGBoost, LightGBM, CatBoost

Example Scenario

Feature	Value
Age	45
Credit Score	750
Geography	Germany
Gender	Female
Active Member	Yes

etc..

Predicted Churn Probability: 76%

Contributing

Contributions are welcome!

Steps to contribute:

1. Fork the repository
2. Create your feature branch
3. Commit changes
4. Push to the branch
5. Create a pull request

---

Developed with ❤️ by Sayed Gamal