Credit Card Default Risk Analysis

This project aims to predict credit card default risk based on historical customer data. We use machine learning techniques to build a classification model that identifies whether a customer will default on their payment in the next month.

Project Description

The dataset used in this project is sourced from the UCI Machine Learning Repository. It contains information about credit card clients, such as credit limit, age, gender, payment history, and bill amounts. The goal is to predict the default payment next month variable, which indicates whether a customer will default (1) or not (0).

The project follows these steps:

1. Data Import and Exploration: Loading and initial analysis of the dataset.
2. Data Cleaning and Preprocessing: Handling missing values, encoding categorical variables, and normalizing numeric features.
3. Modeling: Training a RandomForestClassifier to predict default risk.
4. Model Evaluation: Using metrics like precision, recall, F1-score, and confusion matrix to evaluate model performance.
5. Visualization: Generating charts to interpret feature importance and the confusion matrix.

Technologies Used

• Python: Primary programming language.
• Pandas: Data manipulation and analysis.
• NumPy: Numerical computations.
• Scikit-learn: Machine learning (modeling, cross-validation, metrics).
• Matplotlib and Seaborn: Data visualization.
• Joblib: Saving and loading the trained model.

Project Structure

credit_card_analysis/
├── credit_card_analysis.py       # Main analysis script
├── default of credit card clients.xls  # Dataset
├── credit_card_default_model.pkl # Saved trained model
├── README.md                     # Project documentation
├── requirements.txt              # Project dependencies
├── log.txt                       # Log file for tracking script execution
├── images/                       # Folder containing visualizations
│   ├── feature_importance.png    # Feature importance plot
│   └── confusion_matrix.png      # Confusion matrix plot
│   └── others                    # Other images
└── venv/                         # Virtual environment folder (optional)

Results

The model achieved an average accuracy of 81.7% with cross-validation. However, the recall for class 1 (defaulters) was 36%, indicating that the model struggles to correctly identify customers who will default. Below are some visualizations generated:

1. Feature Importance:

   

2. Confusion Matrix:

   

3. Others:

Developer Info

. Developer: Edson Copque
. Website: https://linktr.ee/edsoncopque
. GitHub: https://github.com/ecopque
. Signal Messenger: ecop.01