🏦 AI-Powered Bank Customer Churn Prediction & Retention Intelligence System

An end-to-end Machine Learning + Business Intelligence project that predicts bank customer churn and provides actionable retention recommendations through a deployed Streamlit web application.

🔗 Live Demo: https://bankchurnretentionsystem-dn4cosylzd2y72yrse22h6.streamlit.app/

📂 GitHub Repository: https://github.com/Gitprem2004/Bank_Churn_Retention_System

---

📌 Project Overview

Customer churn is a major challenge for banks because losing existing customers directly affects revenue and customer lifetime value. This project builds a Churn Prediction & Retention Intelligence System that:

• Predicts churn probability for individual customers.
• Identifies high-risk customers proactively.
• Prioritizes retention actions.
• Recommends next-best actions and customer offers.
• Provides a business-friendly dashboard for decision-making.

The solution combines machine learning, feature engineering, threshold optimization, and deployment engineering into a complete production-style workflow.

---

🎯 Business Objective

• Reduce customer churn.
• Improve retention campaign effectiveness.
• Enable proactive customer engagement.
• Support business teams with interpretable risk insights.

---

🧠 Machine Learning Workflow

1. Data Cleaning
2. Exploratory Data Analysis (EDA)
3. Feature Engineering
4. Feature Selection
5. Train-Test Split
6. Feature Scaling
7. Model Training
8. Model Comparison
9. Cross-Validation
10. Hyperparameter Tuning
11. Threshold Optimization
12. Retention Intelligence Integration
13. Streamlit Deployment

---

🏗️ Solution Architecture

Customer Data
      ↓
EDA & Feature Engineering
      ↓
Feature Selection
      ↓
Train-Test Split
      ↓
StandardScaler
      ↓
Logistic Regression
      ↓
Threshold Optimization (0.30)
      ↓
Retention Intelligence Engine
      ↓
Streamlit Web Application

---

📊 Dataset

The dataset includes customer demographic and banking behavior features such as:

• Age
• Monthly Income
• Years with Bank
• Debit Amount
• Credit Amount
• Debit Transaction Count
• Credit Transaction Count
• Gender
• Marital Status

Target Variable: Status (1 = Churn, 0 = Active)

---

⚙️ Models Evaluated

Model	Accuracy	Precision	Recall	F1-score
Logistic Regression	0.856	0.636	0.467	0.539
Decision Tree	0.804	0.452	0.422	0.437
Random Forest	0.840	0.593	0.356	0.444
Random Forest (Balanced)	0.848	0.640	0.356	0.457

✅ Final Selected Model

Baseline Logistic Regression

Chosen because it achieved the best overall balance between Precision, Recall, F1-score, interpretability, and business usability.

---

🔍 Threshold Optimization

Instead of using the default probability threshold of 0.50, multiple thresholds were tested.

Final Business Threshold: 0.30

This improved churn capture and made the system more useful for retention teams.

• Probability ≥ 0.30 → High Churn Risk
• Probability < 0.30 → Low Churn Risk

---

💡 Retention Intelligence System

The deployed application generates business recommendations based on churn probability.

Risk Levels

• 🔴 High Risk
• 🟠 Medium Risk
• 🟡 Low Risk
• 🟢 Stable Customer

Example Actions

• Immediate retention call
• Personalized offer
• Loyalty program enrollment
• Engagement monitoring
• Relationship manager follow-up

---

🖥️ Streamlit Dashboard Features

• Customer profile input form
• Transaction summary
• Churn probability prediction
• Risk classification
• SLA prioritization
• Next-best-action recommendation
• Offer recommendation
• Downloadable retention report

Dashboard Preview

🚀 Live Application

🔗 Public Streamlit App https://bankchurnretentionsystem-dn4cosylzd2y72yrse22h6.streamlit.app/

---

📁 Project Structure

Bank_Churn_Retention_System/
│
├── app.py
├── requirements.txt
├── README.md
├── .gitignore
│
├── models/
│   ├── churn_model.pkl
│   ├── scaler.pkl
│   └── feature_names.pkl
│
├── notebook/
│   └── Bank_Churn_Analysis.ipynb
│
├── presentation/
│   └── Bank_Churn_Senior_ML_Client_Presentation_Jii.pptx
│
├── images/
│   ├── dashboard_home.png
│   ├── dashboard_analysis.png
│   └── architecture.png
│
└── data/
    └── sample_data.csv

---

🛠️ Tech Stack

Programming: Python ML: scikit-learn Data: pandas, NumPy Deployment: Streamlit Cloud Model Serialization: joblib Version Control: Git & GitHub

---

▶️ Run Locally

Clone Repository

git clone https://github.com/Gitprem2004/Bank_Churn_Retention_System.git
cd Bank_Churn_Retention_System

Install Dependencies

pip install -r requirements.txt

Run Application

streamlit run app.py

---

📈 Key Business Impact

• Earlier churn detection
• Focused retention campaigns
• Reduced unnecessary outreach
• Better customer prioritization
• Actionable retention intelligence
• Improved decision support for business teams

---

🧩 Challenges Faced

• Class imbalance handling
• Threshold selection for business use case
• Feature consistency during deployment
• Model loading path issues
• Streamlit cloud dependency management

Each issue was resolved and documented during the project lifecycle.

---

🔮 Future Enhancements

• XGBoost / LightGBM experimentation
• SHAP explainability
• CRM integration
• Real-time API deployment
• Automated retention workflow engine

---

📚 What I Learned

This project helped me understand the complete machine learning lifecycle:

• Data preprocessing
• Feature engineering
• Model evaluation
• Business interpretation
• Threshold tuning
• Deployment preparation
• Public cloud deployment
• Technical documentation
• Stakeholder communication

---

👨‍💻 Author

Premkumar S B.Tech – Artificial Intelligence & Data Science Chennai, India

• GitHub: https://github.com/Gitprem2004
• LinkedIn: https://www.linkedin.com/in/prem19

---

⭐ If you found this project useful

Please consider giving this repository a star ⭐ on GitHub.