Credit Risk Modeling

Overview

This project focuses on building a credit risk prediction model to identify whether a loan applicant is likely to default or not.

The objective is to support data-driven lending decisions and reduce default-related financial risk.

Dataset

The dataset contains loan application records with:

• Applicant demographic information
• Financial attributes
• Credit history indicators

Target variable:

• loan_status = 1 → Default
• loan_status = 0 → Non-default

Approach

The following models were developed and compared:

• Baseline Logistic Regression
• Logistic Regression with Random Oversampling (ROS)
• Logistic Regression with SMOTE

Class imbalance was explicitly handled to improve default detection.

Model Evaluation

Models were evaluated using:

• Precision & Recall (Default class)
• Macro-averaged metrics
• ROC–AUC score

In credit risk problems, recall for defaulters is prioritised to minimise financial losses.

Results

• Baseline model was conservative and missed many defaulters
• Random oversampling improved recall but increased false positives
• SMOTE provided the most balanced performance

Tools & Libraries

• Python
• Pandas, NumPy
• Scikit-learn
• Imbalanced-learn

Conclusion

The SMOTE-based logistic regression model provided the best balance between risk detection and overall stability, making it suitable for practical credit risk assessment.