Linear Discriminant Analysis in Credit Scoring: A Transparent Hybrid Model Approach

Simpler models can be as effective as complex ones: The research demonstrates that using Linear Discriminant Analysis to reduce features didn't harm performance—the hybrid model achieved 99.45% accuracy while being less computationally demanding, suggesting that transparency doesn't require sacrificing predictive power in credit scoring.

Explainability is essential for fair lending decisions: The paper emphasizes that credit scoring "places importance on decision fairness" and applies two different explainable AI techniques (LIME and Morris Sensitivity Analysis) to interpret model decisions, highlighting that stakeholders need to understand why credit is approved or denied.

Feature reduction improves both efficiency and interpretability: By reducing the number of input features through LDA, the model became less complex while maintaining high performance, making it easier to explain decisions and reducing computational costs—particularly valuable for organizations with limited resources.

Multiple explanation methods provide better transparency: The use of both local explanations (LIME) and global sensitivity analysis (Morris) suggests that different stakeholders may need different types of explanations, and combining approaches offers more comprehensive insight into how the model makes decisions.

Hybrid approaches balance competing priorities: The XG-DNN hybrid model demonstrates that combining different machine learning techniques can achieve the "sweet spot" between accuracy, interpretability, and computational efficiency—all critical for responsible credit scoring systems.