نرگس مهرابی هشتچین

This study developed a data-driven framework for financial statement fraud detection by benchmarking machine learning, deep learning, and hybrid classifiers under a unified, leakage-resistant evaluation protocol. The fraud cases were identified from the U.S. Securities and Exchange Commission’s Accounting and Auditing Enforcement Releases (AAERs) and matched with Compustat data over 1991–2014, producing 122,526 firm-year observations, including 902 confirmed fraud cases. Four structured-input configurations were evaluated: 28 raw financial statement items, 14 financial ratios, their combined set (28+14), and a parsimonious seven-feature subset (six ratios plus Altman’s Z-score). The features were selected using minimum redundancy–maximum relevance (mRMR), class imbalance was addressed via cost-sensitive learning, and performance was assessed with a firm-level 80/20 split and stratified group-based five-fold cross-validation within training. The empirical results indicated that deep and hybrid models consistently outperform classical tabular baselines, reflecting non-linear and interaction-driven fraud signals. The Transformer achieved the most stable and highest overall performance, reaching 0.98898 accuracy and a 0.51087 F1-score under the seven-feature configuration. The combined raw-item and ratio inputs outperformed the ratios alone, implying incremental predictive value in raw accounting items, while the best overall outcomes were obtained with parsimonious seven-feature subset. Collectively, the findings supported the study’s hypotheses and demonstrated the effectiveness of attention-based modeling for financial statement fraud detection.