Bank Customer Churn Prediction Using Artificial Neural Networks

Posted on 2025-01-22

Tags: Python, TensorFlow, Keras, ML Classification, ANN, Feature Engineering

Problem:

A mid-sized bank wanted to proactively identify customers at high risk of churn so that retention strategies could be implemented before accounts were closed. The goal was to build a predictive model that could accurately forecast churn using historical banking data.

Approach:

  • Cleaned and explored the dataset, which included customer demographics, account activity, and tenure
  • Performed feature engineering, including:
    • One-hot encoding for categorical variables
    • Scaling and normalization for continuous features
  • Built an Artificial Neural Network (ANN) using TensorFlow/Keras with:
    • Input layer for engineered features
    • Hidden layers using ReLU activation
    • Sigmoid output for binary classification (churn vs. retain)
  • Evaluated model performance using accuracy, precision, and recall

Outcome:

  • Achieved strong model performance with balanced accuracy and recall across churn classes
  • Enabled segmentation of customers by churn probability for targeted retention campaigns
  • Demonstrated how ANN models can outperform traditional classifiers when nonlinear relationships are present in customer behavior data

Repo: GitHub - Bank Customer Churn Prediction
Demo: (Add Jupyter Notebook render or dashboard link if available)