ML-Based Customer Risk Analysis for Banking: Achieving 80% Recall on Imbalanced Data

Project Overview

This project focuses on customer risk prediction in a banking environment using machine learning models. The primary challenge was dealing with imbalanced data where the minority class represents customers with a higher risk of default. Our goal was to achieve high recall, specifically targeting 80%, while maintaining acceptable levels of precision and accuracy.

Objectives

• Goal: Achieve 80% recall on high-risk customers.
• Dataset: Imbalanced banking dataset.
• Models Used: CART, Random Forest (RF), Gradient Boosting Machine (GBM), LightGBM, and BalancedRandomClassifier.
• Evaluation Metrics: Accuracy, Precision, Recall, F1-Score.

Approach

1. Data Preprocessing:

   • Handled missing data.
   • Performed feature scaling and encoding for categorical variables.
   • Addressed the class imbalance using specialized techniques.

2. Imbalanced Data Handling:

   • Implemented RandomUnderSampler to reduce majority class size.
   • Used TomekLinks to remove overlapping data points and further refine the dataset.

3. Modeling:

   • Tried several machine learning models: CART, RF, GBM, and LightGBM.
   • The most effective model for handling imbalance was BalancedRandomClassifier.

4. Hyperparameter Optimization:

   • Applied hyperparameter tuning to the BalancedRandomClassifier using grid search to optimize performance.

5. Model Evaluation:

   • The BalancedRandomClassifier provided the best results.

Dataset

The dataset used in this project can be downloaded from Kaggle:

Credit Card Approval Prediction Dataset

• Customer Demographics: Age, gender, occupation, etc.
• Financial Indicators: Credit history, balance, transaction patterns.
• Target Variable: Customer risk level.

For privacy reasons, the dataset is not included in this repository.

Results

• Best Model: BalancedRandomClassifier
• Final Performance Metrics:
  • Accuracy: 74%
  • Precision: 72%
  • Recall: 80%
  • F1-Score: 76%

How to Run

Prerequisites

Install the necessary dependencies by running:

pip install -r requirements.txt

Running the Code

1. Clone the repository:

git clone https://github.com/aysecnkci/banking-risk-analysis-imbalanced-data.git

2. Run the Jupyter notebook to preprocess the data, train the model, and evaluate it:

jupyter notebook risk_analysis_banking_imbalanced.ipynb

Repository Structure

├── README.md
├── requirements.txt
├── notebooks/
│   └── risk_analysis_banking_imbalanced.ipynb

Future Work

• Experiment with deep learning models to improve recall.
• Further tune hyperparameters to explore better performance.

License

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

Acknowledgments

Thanks to the contributors and the machine learning community for resources and support.