🏦 Bank Customer Churn Prediction

Predict which bank customers are about to leave — before they do.
End-to-end ML project with EDA, feature engineering, 3 models, and a live interactive app.

🔴 Try the Live App →

---

📌 Problem Statement

A European bank is experiencing customer churn. Every lost customer means lost revenue — the goal is to proactively identify at-risk customers so the retention team can intervene in time.

Input: Customer profile (demographics, account info, activity)
Output: Churn probability + risk tier + personalised retention recommendations

---

🎬 App Preview

Prediction Tab	Model Performance	Data Insights
Churn risk gauge + key factors	ROC curves + confusion matrices	Churn by geography, age, products

👆 Open the live app — no setup needed

---

📊 Dataset

• Source: Churn Modelling — Kaggle
• Size: 10,000 customers from France, Germany, Spain
• Target: Exited (1 = churned, 0 = retained)
• Class split: ~80% retained / ~20% churned

Feature	Description
CreditScore	Customer's credit score
Geography	Country (France / Germany / Spain)
Gender	Male / Female
Age	Customer age
Tenure	Years as a customer
Balance	Account balance
NumOfProducts	Number of bank products held
HasCrCard	Has a credit card (0/1)
IsActiveMember	Active in last 6 months (0/1)
EstimatedSalary	Estimated annual salary

---

🔬 Project Workflow

Data Loading → EDA → Feature Engineering → Preprocessing → Model Training → Evaluation → Deployment

1️⃣ Exploratory Data Analysis

• Class imbalance analysis (80/20 split)
• Distribution plots for all numeric features, split by churn status
• Churn rate by geography, gender, products, and activity
• Correlation heatmap
• Age × Geography churn heatmap

2️⃣ Feature Engineering

Four new features engineered on top of the base dataset:

Feature	Logic	Intuition
BalancePerProduct	Balance / (NumOfProducts + 1)	Wealth concentration per product
AgeGroup	Binned age (18–30, 31–40, ...)	Non-linear age effect
IsHighBalance	Balance > median	Binary wealth flag
TenurePerAge	Tenure / Age	Loyalty relative to lifetime

3️⃣ Models Trained

Model	Accuracy	ROC-AUC	F1 Score	CV AUC
🌲 Random Forest	~86%	~88%	~74%	~87%
📈 Gradient Boosting	~86%	~88%	~73%	~87%
📉 Logistic Regression	~81%	~83%	~60%	~83%

✅ Best model: Random Forest — deployed in the Streamlit app

4️⃣ Key Findings

• 🇩🇪 Germany has the highest churn rate (~32%) vs France (~16%)
• 👴 Customers aged 40–60 churn significantly more than younger ones
• 😴 Inactive members are ~2× more likely to churn
• 📦 Customers with 3–4 products show >80% churn rate (likely over-sold)
• 💳 Having a credit card alone shows no significant churn difference

---

🚀 Run Locally

# Clone
git clone https://github.com/IshanGupta09/Bank_Customer_Churn_Prediction.git
cd Bank_Customer_Churn_Prediction

# Install
pip install -r requirements.txt

# Launch app
streamlit run app.py

Opens at http://localhost:8501

---

📁 Repository Structure

Bank_Customer_Churn_Prediction/
│
├── app.py                              # Streamlit application (live demo)
├── Bank_Customer_Churn_Prediction.ipynb  # Full analysis notebook
├── requirements.txt                    # Python dependencies
├── .gitignore                          # Python + Streamlit gitignore
├── LICENSE                             # MIT
└── README.md                           # You are here

---

🛠 Tech Stack

Category	Tools
Language	Python 3.10+
ML	Scikit-learn (Random Forest, Gradient Boosting, Logistic Regression)
Data	Pandas, NumPy
Visualisation	Plotly, Matplotlib, Seaborn
App	Streamlit
Deployment	Streamlit Cloud

---

💡 Business Impact

Assuming 10,000 customers and $1,200 average annual revenue per customer:

• Current churn (~20%) = $2.4M revenue at risk / year
• Model catches ~60% of churners; retention rate of 40%
• Estimated $576K annual revenue saved through model-driven interventions

---

👨‍💻 Author

Ishan Gupta — CS Engineer specialising in Big Data Analytics

---

⭐ If you found this useful, consider starring the repo!