Self-adaptive MOEA feature selection for classification of bankruptcy prediction data

Area of Science:

• Finance and Accounting
• Computational Intelligence

Background:

• Assessing bankruptcy risk is crucial for creditors and investors.
• Increasing company complexity and sophisticated financial reporting make bankruptcy prediction challenging.
• Traditional methods struggle with high-dimensional datasets and irrelevant features.

Purpose of the Study:

• To develop a novel methodology for feature selection in bankruptcy data classification.
• To employ an evolutionary multiobjective approach for optimizing feature subsets and classifier performance.
• To integrate self-adaptation for simultaneous feature selection and classifier parameter optimization.

Main Methods:

• Utilized an evolutionary multiobjective algorithm for feature selection.
• Simultaneously minimized the number of features and maximized classifier accuracy.
• Applied self-adaptation to optimize classifier parameters concurrently with feature selection.
• Tested the methodology on four diverse bankruptcy datasets.

Main Results:

• The proposed methodology effectively performed feature selection for bankruptcy data classification.
• Self-adaptation of the classifier parameters alongside feature selection yielded improved results.
• Demonstrated the utility of the integrated approach in enhancing bankruptcy prediction models.

Conclusions:

• The evolutionary multiobjective approach with self-adaptation is effective for bankruptcy prediction.
• This methodology enhances the accuracy and efficiency of credit risk assessment.
• The approach offers a robust solution for complex financial data analysis.