Credit Risk Assessment: Loan Default Prediction

Predicting Creditworthiness Using Machine Learning

Author: Jamiu Olamilekan Badmus
Email: jamiubadmus001@gmail.com
LinkedIn: Jamiu Olamilekan Badmus
GitHub: jamiubadmusng
Website: sites.google.com/view/jamiu-olamilekan-badmus

Executive Summary

This project develops a machine learning model to assess credit risk and predict loan defaults. Using the German Credit dataset, we build classification models that help financial institutions make informed lending decisions while managing risk exposure.

Key Results:

Built and evaluated 6 classification models on 1,000 credit applicants
Engineered risk-based features from 20 applicant attributes
Implemented cost-sensitive classification for business optimization
Achieved strong predictive performance with interpretable SHAP analysis

Problem Statement

Credit risk assessment is fundamental to banking operations:

$1.5 trillion in consumer loans are issued annually in the US
Default rates of 2-5% can significantly impact profitability
Regulatory requirements (Basel III) mandate robust risk assessment
Automated scoring enables faster, more consistent decisions

This project addresses: How can we predict which loan applicants are likely to default, enabling risk-adjusted lending decisions?

Business Costs

In credit risk, different types of errors have asymmetric costs:

False Negative (approving a defaulter): Loss of principal + interest
False Positive (rejecting a good customer): Lost profit opportunity

Typically, the cost of a missed default is 5x the cost of a lost opportunity.

Data Source

The dataset is the German Credit Dataset from the UCI Machine Learning Repository.

Dataset Characteristics

Attribute	Value
Source	UCI ML Repository
Observations	1,000 credit applicants
Features	20 (7 numerical, 13 categorical)
Target	Good Credit (70%) / Bad Credit (30%)
Time Period	Historical German bank data

Key Features

Feature	Description
checking_status	Status of existing checking account
duration	Loan duration in months
credit_history	Past credit behavior
purpose	Purpose of the loan
credit_amount	Loan amount requested
savings_status	Savings account balance
employment	Present employment duration
age	Age of applicant
housing	Housing situation (rent/own)
job	Job type and skill level

Project Structure

finance/
├── data/
│   ├── raw/                          # Original dataset
│   │   └── german_credit.data
│   └── processed/                    # Feature-engineered data
│       └── german_credit_processed.csv
├── docs/
│   ├── analysis_report.md            # Detailed analysis write-up
│   └── figures/                      # Visualization outputs
│       ├── target_distribution.png
│       ├── numerical_distributions.png
│       ├── categorical_default_rates.png
│       ├── correlation_matrix.png
│       ├── model_comparison.png
│       ├── confusion_matrix.png
│       ├── roc_pr_curves.png
│       ├── threshold_optimization.png
│       ├── shap_importance.png
│       ├── shap_beeswarm.png
│       └── risk_distribution.png
├── models/                           # Trained model artifacts
│   ├── credit_risk_model.joblib
│   └── preprocessor.joblib
├── notebooks/
│   └── credit_risk_assessment.ipynb  # Main analysis notebook
├── src/
│   └── predict_risk.py               # Standalone Python module
├── README.md                         # This file
├── requirements.txt                  # Python dependencies
├── LICENSE                           # MIT License
└── .gitignore                        # Git ignore file

Installation

Prerequisites

Python 3.8 or higher
pip package manager

Setup

Clone the repository:

git clone https://github.com/jamiubadmusng/credit-risk-assessment.git
cd credit-risk-assessment

Create and activate a virtual environment:

python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

Install dependencies:
```
pip install -r requirements.txt
```

Usage

Running the Jupyter Notebook

cd notebooks
jupyter notebook credit_risk_assessment.ipynb

Using the Trained Model

import joblib
import pandas as pd

# Load model and preprocessor
model = joblib.load('models/credit_risk_model.joblib')
preprocessor = joblib.load('models/preprocessor.joblib')

# Prepare applicant data
applicant = pd.DataFrame({
    'checking_status': ['A11'],
    'duration': [24],
    'credit_history': ['A32'],
    # ... other features
})

# Preprocess and predict
X_processed = preprocessor.transform(applicant)
default_probability = model.predict_proba(X_processed)[:, 1]
print(f"Default Probability: {default_probability[0]:.2%}")

Methodology

1. Data Preprocessing

Decoded categorical variables for interpretability
Converted target variable (1=Good → 0, 2=Bad → 1)
Created derived features (credit-to-income ratio, monthly payment)

2. Feature Engineering

Numerical Features:

Duration, credit amount, installment rate
Age, residence duration, existing credits

Categorical Features:

Checking account status, credit history
Employment, housing, job type

Derived Features:

Credit-to-income ratio
Monthly payment estimate
Stability indicators (employment, housing)

3. Model Training

6 classification algorithms evaluated
5-fold stratified cross-validation
80/20 train-test split

4. Cost-Sensitive Optimization

Implemented asymmetric cost matrix
Optimized classification threshold
Balanced precision-recall for business needs

Results

Model Performance

Model	CV ROC-AUC	Test ROC-AUC	Test F1
Logistic Regression	0.7682	0.8043	0.6055
Random Forest	0.7831	0.8037	0.5263
LightGBM	0.7900	0.7793	0.5766
Gradient Boosting	0.7767	0.7835	0.5636
XGBoost	0.7664	0.7512	0.5546
Decision Tree	0.6753	0.7090	0.4348

Best Model: Logistic Regression with Test ROC-AUC of 80.43%

Cost-Sensitive Analysis

Default threshold (0.5): Total Cost = 151
Optimal threshold (0.30): Total Cost = 93
Cost reduction: 38.4%

Threshold	FP	FN	Recall	Total Cost
Default (0.50)	16	27	55.0%	151
Optimal (0.30)	38	11	81.7%	93

Key Findings

1. Checking Account Status is Critical

Applicants without a checking account or with low balance show significantly higher default rates. This is the strongest single predictor.

2. Loan Duration Matters

Longer loan durations correlate with higher default risk. Short-term loans (< 12 months) have lower default rates.

3. Credit History Drives Risk

Past payment behavior strongly predicts future defaults. Critical accounts have 2-3x higher default rates.

4. Cost-Sensitive Threshold Improves Business Outcomes

Optimizing the classification threshold for business costs reduces total expected loss.

Business Recommendations

Risk Segmentation Strategy

Risk Level	Probability	Actual Default Rate	Action
Very Low	< 20%	11.0%	Auto-approve
Low	20-40%	17.1%	Standard approval
Medium	40-60%	54.5%	Enhanced review
High	60-80%	70.8%	Decline or collateral
Very High	> 80%	72.7%	Decline

Implementation Roadmap

Immediate: Deploy model for new application scoring
Short-term: Implement tiered interest rates by risk category
Long-term: Integrate with loan management system

Future Work

Alternative Data: Incorporate transaction history, social data
Model Monitoring: Implement drift detection and retraining
Fairness Analysis: Audit for demographic bias
Deep Learning: Explore neural networks for complex patterns

References

UCI Machine Learning Repository - German Credit Dataset
Basel Committee on Banking Supervision - Basel III Framework
Lundberg, S. M., & Lee, S. I. (2017). A Unified Approach to Interpreting Model Predictions

License

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

Contact

For questions or collaboration opportunities:

Email: jamiubadmus001@gmail.com
LinkedIn: Jamiu Olamilekan Badmus
GitHub: jamiubadmusng
Website: sites.google.com/view/jamiu-olamilekan-badmus

Name		Name	Last commit message	Last commit date
data		data
docs		docs
models		models
notebooks		notebooks
src		src
.gitattributes		.gitattributes
.gitignore		.gitignore
LICENSE		LICENSE
README.md		README.md
requirements.txt		requirements.txt

Latest commit

History

Folders and files

Repository files navigation