Masumi Nakashima

Bunkyo Gakuin University, Tokyo, Japan

ABSTRACT

This study provides a systematic analysis of machine learning-based financial statement fraud detection research using a systematic literature review, science mapping, and text-mining techniques. Based on 85 peer-reviewed articles published between 2009 and 2023, the study examines the intellectual structure, dominant themes, and evolution of the field. The findings reveal a shift from traditional statistical models to advanced machine learning and deep learning approaches, together with the increasing use of unstructured data such as textual disclosures. Major research themes include model development, feature selection, performance evaluation, and data integration. Despite methodological advances, the literature shows limited integration with fraud-related theories, including the fraud triangle and fraud diamond frameworks, reducing model interpretability and practical applicability. The study highlights challenges related to explainability, data imbalance, and ethical concerns, while emphasizing the need for stronger theoretical foundations. By synthesizing fragmented findings and linking data-driven methods with accounting theory, this study contributes to a deeper understanding of fraud detection research and offers implications for auditing practice, corporate governance, regulatory oversight, and future research.

Keywords:
Financial Statement Fraud
Machine Learning
Fraud Detection
Science Mapping
Text Mining
Systematic Literature Review