Integrating the fraud triangle with machine learning for financial misstatement detection: Evidence from an emerging market

This study explores integrating the Fraud triangle theory with machine-learning-based data analytics to enhance the detection of material financial statement misstatements in an emerging market context. Using 6,669 firm-year observations from Vietnamese publicly listed firms between 2019 and 2023, the research applies Logistic regression, Random forest, and Extreme gradient boosting (XGBoost) models to classify material misstatements, defined as a 10% deviation in profit under Vietnamese Auditing Standard No. 320. The class imbalance problem was addressed using the Synthetic Minority Oversampling Technique (SMOTE), and Bayesian Optimization was employed to fine-tune hyperparameters for optimal model performance. The results demonstrate that ensemble learning models, particularly Random Forest, outperform traditional statistical methods, achieving the highest accuracy (0.8403), F1-score (0.8037), and AUC (0.9175). These findings emphasize the superior predictive power of machine learning in uncovering nonlinear fraud patterns that conventional audits may miss. The study contributes to the auditing and accounting literature by bridging behavioural theory and technological innovation, providing empirical insights from an emerging market. The findings also offer practical implications for auditors, regulators, and policymakers aiming to strengthen corporate transparency and audit quality through data-driven assurance systems.