THE POWER OF MACHINE LEARNING MODELS IN EARLY DETECTION OF FRAUDULENT FINANCIAL STATEMENTS: THE EXAMPLE OF BORSA ISTANBUL

Abstract

Detecting and preventing fraudulent financial statements is crucial to maintaining the reliability of financial markets, as such statements undermine stakeholders and disrupt healthy market functioning. This study develops an artificial intelligence–supported model to detect fraudulent financial statements of companies listed on Borsa Istanbul. Using financial and non-financial data, Random Forest, Naive Bayes, and K-Star classification analyses were applied to firms identified as preparing fraudulent statements according to Capital Markets Board bulletins between 01.01.2022 and 01.01.2025. Results indicate that the K-Star algorithm achieved 99% accuracy in detecting fraudulent statements one period in advance, compared with 92% for Random Forest and 62% for Naive Bayes. Specifically, K-Star classified fraudulent firms with 100% accuracy and non-fraudulent firms with 98% accuracy. These results highlight that combining financial and non-financial indicators offers a novel and effective approach to fraud detection and provides significant contributions to the literature.