AI-Driven Fraud Detection in Financial Statements: A Comparative Study of Machine Learning Models

Abstract

The increasing complexity of financial transactions and the growing volume of corporate data have significantly intensified the risk of financial statement fraud, challenging the effectiveness of traditional detection methods. This study examines the role of artificial intelligence in enhancing fraud detection by conducting a comparative analysis of machine learning and deep learning models applied to financial statement data. Drawing on an extensive review of prior literature and empirical findings, the research evaluates the performance of widely used algorithms, including logistic regression, decision trees, support vector machines, random forest, gradient boosting, and deep learning architectures. The results indicate that ensemble and deep learning models consistently outperform conventional statistical and rule-based approaches in terms of accuracy, precision, and overall predictive capability. However, the findings also highlight critical challenges related to data imbalance, model interpretability, and computational complexity, which may limit the practical adoption of advanced models in real-world auditing and regulatory contexts. The study contributes to the existing literature by providing a structured comparison of AI-driven fraud detection techniques and offers practical implications for auditors, regulators, and policymakers seeking to improve the reliability and transparency of financial reporting. Future research directions include the development of explainable and hybrid models, the integration of non-financial and textual data, and the expansion of cross-country datasets to enhance model generalizability and support sustainable capital market development.