Fraud Risk and Audit Opinions Across Countries: Complementing Accounting-Based Fraud Risk with Machine Learning Methods

Financial statement fraud is a significant threat to corporate governance, investor confidence, and global capital markets. Traditional fraud detection models, including DF-Score and PF-Score, often rely on linear approaches that may fail to capture complex fraudulent behaviors. This study applies machine learning techniques using eXtreme Gradient Boosting across 34 countries (2016–2024). Results show that enhanced DF-Score and PF-Score effectively capture fraud risk, which is significantly associated with auditors’ opinions. The study integrates machine learning with traditional models to address data complexity and nonlinearity. Practically, the findings provide auditors, regulators, and financial analysts with a tool for improved fraud detection and risk-based auditing.