Artificial Intelligence and Data Mining in Detecting Financial Statement Fraud: A Systematic Literature Review

General Background: Fraud in financial reporting significantly undermines stakeholder confidence and destabilises financial markets. Specific Background: The increasing complexity of financial data makes traditional fraud detection techniques inadequate, necessitating more sophisticated methods such as data mining and artificial intelligence (AI). Knowledge Gap: Despite the increasing adoption of AI in fraud detection, previous systematic literature reviews (SLRs) have generally focused narrowly on specific algorithms or data types, thus failing to provide a comprehensive assessment across multiple contexts. Objective: This study aims to critically evaluate the application of AI and data mining techniques in detecting financial statement fraud through a systematic literature review. Methods: A total of 30 peer-reviewed articles published between 2014 and 2024 were selected from Scopus, ScienceDirect, and Emerald databases using predefined inclusion-exclusion criteria and analysed narratively. Results: The review identified that supervised learning algorithms, specifically Support Vector Machine (SVM), Logistic Regression (LR), and XGBoost, were predominantly used, with XGBoost (96.94%) and LSTM (94.98%) showing the highest accuracy. Integration of financial and non-financial data improves detection stability. Novelty: In contrast to previous systematic reviews, this study offers a holistic synthesis covering algorithm types, structured and unstructured data, and diverse regional contexts. Implications: The findings highlight the transformative potential of AI in fraud detection and encourage further research on unsupervised learning and more in-depth utilisation of unstructured data