Financial Fraud Identification Based on Stacking Ensemble Learning Algorithm: Introducing MD&A Text Information.

Area of Science:

• Financial analysis
• Machine learning
• Text analytics

Background:

• Increasing incidents of financial fraud necessitate improved detection methods.
• Maintaining capital market order is a key concern for researchers and practitioners.
• Existing financial fraud identification models may lack comprehensive variable integration.

Purpose of the Study:

• To develop an advanced financial fraud identification model.
• To evaluate the efficacy of stacking ensemble learning in fraud detection.
• To assess the contribution of text-based variables to model performance.

Main Methods:

• Construction of a financial fraud identification model using stacking ensemble learning.
• Integration of financial and non-financial variables.
• Inclusion of text variables: sentiment polarity, emotional tone, and readability from the Management Discussion and Analysis (MD&A) chapter.

Main Results:

• The stacking ensemble learning model significantly outperforms individual classifiers.
• Incorporating text variables demonstrably enhances the model's fraud recognition capabilities.
• The combined approach using financial, non-financial, and text data yields superior results.

Conclusions:

• The developed stacking ensemble model offers a more effective method for financial fraud identification.
• Textual analysis of annual reports provides valuable insights for fraud detection.
• This approach holds promise for improving capital market integrity.