Credit Risk Prediction

Predictive model for credit default risk using machine learning classification techniques.

Overview

This project builds a binary classification model to predict credit default risk using historical loan data. The model achieves 0.95 AUC-ROC score and uses SHAP values for model interpretability, making predictions explainable for business stakeholders.

Dataset

• Source: Kaggle Credit Risk Dataset
• Size: 32,581 loan records
• Features: 12 variables including demographics, loan characteristics, and credit history
• Target: Binary classification (default vs. non-default)
• Class distribution: 21.8% default rate (imbalanced)

Key Results

Metric	Logistic Regression	XGBoost
AUC-ROC	0.87	0.95
Accuracy	0.86	0.92
Precision (Default)	0.75	0.83
Recall (Default)	0.55	0.79

Methodology

1. Data Preprocessing

• Handled missing values using median imputation
• Applied One-Hot Encoding for categorical variables
• Standardized numerical features using StandardScaler
• Created income brackets for analysis

2. Feature Engineering

• Encoded categorical variables (home ownership, loan intent, loan grade)
• Scaled numerical features to comparable ranges
• Addressed class imbalance using scale_pos_weight parameter

3. Model Development

• Baseline: Logistic Regression for comparison
• Main Model: XGBoost Classifier optimized for imbalanced data
• Validation: 80/20 train-test split with stratification

4. Model Interpretability

• SHAP (SHapley Additive exPlanations) for feature importance
• Global interpretation: Which features matter most overall
• Local interpretation: Why individual predictions were made

Key Findings

Most Important Features for Default Prediction:

1. person_income - Higher income strongly reduces default risk
2. loan_percent_income - Higher loan-to-income ratio increases risk
3. loan_int_rate - Higher interest rates correlate with defaults
4. loan_intent_VENTURE - Venture loans show higher risk
5. person_home_ownership - Ownership status affects risk profile

Business Insights:

• Low-income applicants with high loan-to-income ratios are highest risk
• Interest rate is both a predictor and consequence of risk
• Loan grade F/G customers have significantly elevated default probability
• Age shows concentration of defaults in 20-30 year range with high loan amounts

Technologies Used

• Python 3.8+
• Data Processing: pandas, numpy
• Visualization: matplotlib, seaborn
• Machine Learning: scikit-learn, XGBoost
• Interpretability: SHAP
• Environment: Google Colab

Installation & Usage

# Clone repository
git clone [your-repo-url]

# Install dependencies
pip install -r requirements.txt

# Run notebook
jupyter notebook Credit_Risk_Analysis.ipynb

Project Structure

credit-risk-prediction/
├── Credit_Risk_Analysis.ipynb    # Main analysis notebook
├── credit_risk_dataset.csv       # Dataset
├── README.md                      # This file
└── requirements.txt               # Dependencies

Requirements

pandas>=1.3.0
numpy>=1.21.0
matplotlib>=3.4.0
seaborn>=0.11.0
scikit-learn>=1.0.0
xgboost>=1.5.0
shap>=0.40.0

Future Improvements

• Hyperparameter tuning using GridSearchCV
• Test ensemble methods (Random Forest, LightGBM)
• Implement cost-sensitive learning
• Deploy model as REST API
• Add feature selection analysis

License

MIT License

Contact

Gastón Schvartz
Email: schvartz.g@gmail.com
LinkedIn: gaston-schvartz

---

Credit risk analysis project demonstrating end-to-end machine learning workflow from data preprocessing to model interpretability.