🏦 Bank Customer Churn Prediction using Artificial Neural Network (ANN)
Overview
This project is an AI-powered Bank Customer Churn Prediction System developed using Deep Learning (Artificial Neural Networks). The application predicts whether a customer is likely to leave the bank based on their personal and financial information.
The model is deployed using Streamlit and can be easily hosted on Hugging Face Spaces for public access.
Features
- Deep Learning-based customer churn prediction
- Interactive Streamlit web interface
- Real-time predictions
- Standardized input preprocessing using StandardScaler
- Probability score for each prediction
- Clean and user-friendly interface
Dataset
The project uses the Bank Customer Churn Dataset.
Input Features
| Feature | Description |
|---|---|
| CreditScore | Customer's credit score |
| Gender | Male or Female |
| Age | Customer age |
| Tenure | Number of years with the bank |
| Balance | Current account balance |
| NumOfProducts | Number of bank products owned |
| HasCrCard | Whether the customer has a credit card |
| IsActiveMember | Whether the customer is an active member |
| EstimatedSalary | Estimated annual salary |
Target Variable
| Target | Meaning |
|---|---|
| 0 | Customer Stays |
| 1 | Customer Leaves (Churn) |
Technologies Used
- Python
- TensorFlow / Keras
- NumPy
- Pandas
- Scikit-learn
- Joblib
- Streamlit
Deep Learning Architecture
- Input Layer: 9 Features
- Hidden Layer 1: Dense Layer (ReLU)
- Hidden Layer 2: Dense Layer (ReLU)
- Output Layer: Dense Layer (Sigmoid)
Activation Functions
- ReLU
- Sigmoid
Loss Function
Binary Crossentropy
Optimizer
Adam Optimizer
Evaluation Metric
Accuracy
Data Preprocessing
The following preprocessing steps were applied before training:
- Removed unnecessary columns
- Encoded Gender
- Feature Scaling using StandardScaler
- Train-Test Split
- Normalized numerical features
The trained StandardScaler is saved as:
scaler.pkl
to ensure identical preprocessing during deployment.
Model Files
ann_model.h5
Trained Artificial Neural Network model.
scaler.pkl
Saved StandardScaler used during training.
app.py
Streamlit application.
Project Structure
Bank-Customer-Churn-Prediction/
│
├── app.py
├── ann_model.h5
├── scaler.pkl
├── requirements.txt
├── README.md
└── dataset.csv
Installation
Clone the repository
git clone <repository-url>
Move into the project
cd Bank-Customer-Churn-Prediction
Create Virtual Environment
python -m venv venv
Activate Virtual Environment
Windows
venv\Scripts\activate
Linux / macOS
source venv/bin/activate
Install dependencies
pip install -r requirements.txt
Run the Application
streamlit run app.py
How the Prediction Works
- User enters customer information.
- Input data is converted into numerical format.
- Features are scaled using the saved StandardScaler.
- The trained ANN predicts the probability of customer churn.
- If the probability is greater than 0.5, the customer is predicted to leave the bank; otherwise, the customer is predicted to stay.
Example Prediction
Input
Credit Score: 619
Gender: Female
Age: 42
Tenure: 2
Balance: 0
Products: 1
Credit Card: Yes
Active Member: Yes
Estimated Salary: 101348.88
Output
Prediction Probability: 0.34
Customer is likely to Stay.
Requirements
streamlit
tensorflow
numpy
pandas
scikit-learn
joblib
Future Improvements
- Support Geography feature
- Explain predictions using SHAP
- Interactive analytics dashboard
- Batch prediction from CSV
- REST API using FastAPI
- Docker deployment
- Cloud deployment on AWS or Azure
Deployment
This project can be deployed on:
- Hugging Face Spaces
- Streamlit Community Cloud
- Render
- Railway
- AWS EC2
- Microsoft Azure
- Google Cloud Platform
Author
Sudheer Muthyala
B.Tech – Electronics and Communication Engineering
Aspiring Data Scientist | AI & Machine Learning Enthusiast
License
This project is intended for educational and portfolio purposes. Feel free to fork, modify, and build upon it while providing appropriate attribution.