Predicting Customer Churn in Banking: A Neural Network Approach with Enhanced Class Imbalance Techniques

Project Overview

This project aims to predict customer churn in the banking sector using a Feedforward Neural Network. The model leverages enhanced class imbalance techniques to improve prediction accuracy, ensuring better customer retention strategies by identifying customers likely to leave the bank. The insights gained from this model can help banks develop targeted interventions to reduce churn.

Table of Contents

• Project Overview
• Features
• Technologies Used
• Dataset
• Usage
• Results
• Future Improvements
• Contributing
• License

Features

• Feedforward Neural Network for churn prediction.
• Class imbalance handling using techniques such as class weighting to improve prediction accuracy on minority classes.
• Exploratory Data Analysis (EDA) to understand key patterns and insights about customer churn.
• Evaluation metrics: F1 Score, Precision, Recall, and Accuracy.
• Visualization of model performance, including confusion matrix and accuracy trends over epochs.

Technologies Used

• Python
• TensorFlow / Keras
• NumPy, Pandas, Matplotlib, Seaborn

Dataset

The dataset consists of customer information from a bank, including features like age, credit score, balance, number of products, and whether or not they churned (left the bank). The dataset is available from UCI Machine Learning Repository or another source, depending on your dataset source.

Data Fields

• CustomerId: Unique identifier for each customer
• CreditScore: Customer's credit score
• Geography: Customer's country
• Gender: Gender of the customer
• Age: Customer's age
• Tenure: Number of years with the bank
• Balance: Account balance
• NumOfProducts: Number of products held by the customer
• HasCrCard: Whether the customer has a credit card
• IsActiveMember: Whether the customer is an active member
• Exited: Target variable (1 if the customer left the bank, 0 otherwise)

Usage

1. Train the Model: Open the Churn_Prediction_Model.ipynb notebook and follow the steps to preprocess the data, build the model, and train it.
2. Evaluate the Model: After training, evaluate the model performance using the provided evaluation metrics.
3. Visualize Results: The notebook includes sections to visualize model performance (accuracy, precision, recall, and confusion matrix).

Results

• Baseline Model achieved approximately 65% accuracy.
• Enhanced Model with class imbalance handling achieved approximately 67% accuracy.
• The model was able to identify key patterns in customer churn, though further improvements may be required for production-level accuracy.

See the "Results" section in the notebook for a detailed analysis of model performance.

Future Improvements

• Hyperparameter Tuning: Experiment with different architectures, learning rates, and other hyperparameters.
• Additional Features: Incorporate additional data on customer behavior to improve prediction accuracy.
• Advanced Techniques: Explore ensemble methods or transfer learning for better performance.

Contributing

Contributions are welcome! If you'd like to improve the model, add new features, or enhance documentation, please fork the repository and submit a pull request.

License

This project is licensed under the MIT License - see the LICENSE file for details.