Data Driven Approach for Detecting Financial Statement Fraud

(A Systematic Literature Review)

Abstract

Financial statement fraud represents a persistent and complex threat to organizational integrity, requiring more advanced analytical tools to detect subtle accounting manipulations. To provide an evidence-based understanding of how data-driven techniques have been utilized in this domain, this study conducts a Systematic Literature Review (SLR) guided by the PRISMA framework. The review addresses two research questions: (a) what data-driven approaches have been used to detect financial statement fraud, and (b) what are the characteristics of the data, modelling methods, and evaluation metrics employed in prior studies. A structured search and screening process was executed using predefined inclusion and exclusion criteria, enabling the selection of relevant peer-reviewed studies from multiple academic databases. The included articles were further examined using meta-analysis techniques to synthesize quantitative evidence where applicable. The findings reveal that financial statement fraud detection has increasingly shifted toward machine learning, deep learning, graph-based analytics, and other advanced data-driven models capable of identifying hidden or non-linear patterns in financial reporting data. The reviewed studies employ diverse data characteristics, including financial ratios, earnings indicators, transactional records, and graph-structured relationships. Overall, this review highlights both the advancements and the methodological challenges within the field, underscoring the need for improved data quality, consistent evaluation practices, and models that balance predictive performance with interpretability for auditing applications.

 Keywords: Fraud, data-driven, machine learning, anomaly detection, deep learning.