NATE: Non-pArameTric approach for Explainable credit scoring on imbalanced class.

Area of Science:

• Machine Learning
• Financial Analytics
• Data Science

Background:

• Credit scoring models are vital for financial institutions, but traditional methods like logistic regression struggle with complex, imbalanced datasets, impacting accuracy.
• Tree-based models offer better performance but lack interpretability, creating a gap in addressing both predictive power and explainability in credit scoring.
• Class imbalance in credit scoring datasets can significantly degrade model robustness and accuracy, favoring the majority class.

Purpose of the Study:

• To introduce the Non-pArameTric oversampling approach for Explainable credit scoring (NATE) framework.
• To enhance both predictive performance and interpretability in credit scoring models.
• To address challenges posed by imbalanced data distributions and the need for transparent decision-making.

Main Methods:

• NATE framework combines oversampling techniques with tree-based classifiers for improved performance and interpretability.
• Class balancing methods are integrated to mitigate the effects of imbalanced data distributions.
• Interpretability features are incorporated to provide insights into the model's decision-making process.

Main Results:

• NATE significantly outperforms logistic regression in credit risk classification, showing improvements of 19.33% in AUC, 71.56% in MCC, and 85.33% in F1 Score.
• Oversampling with gradient boosting achieved optimal metrics: AUC 0.9649, MCC 0.8104, and F1 Score 0.9072, outperforming undersampling.
• NATE enhances model interpretability by detailing feature contributions, aiding in the understanding of individual predictions.

Conclusions:

• NATE effectively manages class imbalance, boosts predictive performance, and increases model interpretability in credit scoring.
• The framework demonstrates potential as a reliable and transparent tool for credit scoring applications.
• Combining oversampling with gradient boosting offers a powerful approach for accurate and explainable credit risk assessment.