Masumi Nakashima

Bunkyo Gakuin University

ABSTRACT

Forensic accounting has attracted global attention. Machine learning has dramatically advanced fraud detection in accounting. Fraud detection using data mining, AI, and machine learning has become a central topic in forensic accounting, wherein detecting financial statement fraud during auditing is difficult. I extract machine learning financial statement fraud detection studies from 2009 to 2023 using science mapping and identify the most common countries and journals. Second, text-mining techniques such as word frequency, co-occurrence networks, and correspondence analysis through titles and keywords were employed to discover subtopics. This study then identifies current and future trends in recent machine learning financial statement fraud detection modeling studies. Third, this study conducts a meta- analysis to determine which machine learning techniques and evaluations are most used, and the challenges of machine learning financial statement fraud detection research. Systematic mapping identified 56 articles. Since 2015, research using machine learning for financial statement fraud has surged, with the most commonly used method being support vector machines. This study contributes to accounting and legal professionals, investors, regulators, and others interested in financial statement fraud detection research using machine learning.

Keywords: Science Mapping Approach, Machine Learning, Fraud Detection, Meta-Analysis, Text Mining