Hybrid Harmony Search-Artificial Intelligence Models in Credit Scoring.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Computational Finance

Background:

• Credit scoring is crucial for financial institutions to identify defaulters accurately.
• Support Vector Machines (SVM) and Random Forest (RF) are powerful AI models for credit scoring but are sensitive to hyperparameters and act as black-boxes.
• Feature selection and importance computation are key for model interpretability in credit risk assessment.

Purpose of the Study:

• To propose hybrid AI models for enhanced credit scoring accuracy and interpretability.
• To introduce a Modified Harmony Search (MHS) algorithm for efficient hyperparameter tuning and feature selection.
• To evaluate the performance of MHS-hybrid models against standard statistical and AI models.

Main Methods:

• Developed hybrid HS-SVM for simultaneous feature selection and hyperparameter tuning.
• Developed hybrid HS-RF for hyperparameter tuning.
• Proposed MHS with modifications including elitism, dynamic operators, self-adjusted bandwidth, and additional termination criteria for reduced computational time.
• Utilized parallel computing to further decrease computation time.

Main Results:

• The Modified HS-RF model exhibited the most robust performance across three credit scoring datasets.
• MHS-RF achieved superior results in terms of model performance, explainability, and computational efficiency.
• The proposed MHS algorithm significantly reduced the computational time of the hybrid models.

Conclusions:

• Hybrid models, particularly MHS-RF, offer significant improvements in credit scoring accuracy and interpretability.
• The MHS algorithm provides an efficient approach to optimize complex AI models for credit risk analysis.
• The integration of MHS with RF presents a promising direction for developing more transparent and effective credit scoring systems.