Customer Churn Prediction using ANN

This project demonstrates the use of a feed-forward Artificial Neural Network (ANN) to predict customer churn using the IBM Telco Customer Churn dataset. It covers data preprocessing, feature engineering, model training, evaluation, and visualization, providing a complete workflow for churn prediction.

Overview & Objective

• Objective: Predict whether a customer will churn (binary classification).
• Target Variable: Churn column.
• Approach:
  1. Encode categorical features and scale numerical features.
  2. Split dataset into training and test sets.
  3. Train a compact ANN using Keras/TensorFlow.
  4. Evaluate performance using accuracy, confusion matrix, and classification metrics.

Project Structure

• prediction.ipynb — main notebook containing the full workflow.
• WA_Fn-UseC_-Telco-Customer-Churn.csv — dataset used.
• README.md — project documentation.

Dataset

• Source: IBM Telco Customer Churn dataset.
• Description: Includes demographic, service, and billing information of customers, with a Churn label indicating whether the customer left the service.

Model Architecture

• Input Features: 26
• Layers:

  Dense(26, activation='relu', input_shape=(26,))
  Dense(1, activation='sigmoid')

• Compilation:

  • Optimizer: adam
  • Loss: binary_crossentropy
  • Metrics: accuracy

• Training: 100 epochs, validation split 0.2

Preprocessing Steps

1. Categorical Encoding: One-hot encoding for features like InternetService, Contract, PaymentMethod.
2. Numerical Scaling: Min-Max scaling for tenure, MonthlyCharges, TotalCharges.
3. Train-Test Split: 80% training, 20% testing (random_state=5).
4. Input Shape: Ensured consistency with 26 features.

Evaluation

• Metrics Used: Accuracy, Confusion Matrix, Precision, Recall, F1-score.
• Performance: Validation accuracy typically ~0.80.

Environment & Requirements

• Python 3.x
• Install dependencies:

pip install tensorflow scikit-learn pandas numpy matplotlib

How to Run

1. Install dependencies.
2. Open prediction.ipynb in Jupyter Lab, Notebook, or VS Code.
3. Run all cells sequentially to reproduce preprocessing, training, and evaluation.

Notes

• Ensure TotalCharges is numeric before scaling.
• Random seed (random_state=5) ensures reproducibility.
• GPU recommended for faster model training.

Acknowledgements

• IBM Telco Customer Churn dataset
• TensorFlow/Keras and scikit-learn communities

I can also create a **matching `requirements.txt` and GitHub repo description** that complements this README professionally. Do you want me to do that?