The effectiveness of data mining in detecting fraud risk in financial reporting

Fraud in financial reporting occurs when businesses intentionally present misleading information in their financial statements, posing significant risks to investors, regulatory bodies, and the economic system as a whole. Recent studies in the field of accounting have focused on the use of data mining techniques to detect fraud risk. Data mining is a method that employs a range of analytical techniques to discover information and patterns from large datasets. These techniques are notable for their ability to identify unusual patterns and anomalies in financial data. The aim of this thesis is to analyze the financial statement data of companies with potential fraud risk in financial reporting, to elaborate on approaches for detecting fraud risk, and to examine the effectiveness of data mining methods in this process.The research analyzed the annual financial statements of a total of 285 companies listed on Borsa Istanbul from 2017 to 2022, across the Close Monitoring Market, Star Market, and Main Market groups. Companies' markets were considered to classify them as having fraud risk or not. Financial statements of companies listed in the Star and Main Markets were classified as not having fraud risk, while financial statements of companies listed in the Close Monitoring Market were classified as having fraud risk.Nearly all studies related to detecting fraudulent financial reporting rely on financial ratios. Financial ratios are useful not only for assessing a company's past and current financial condition but also for fulfilling auditing and planning functions. Therefore, the use of financial ratios in detecting fraud risk is quite common. In this study, 23 ratios frequently used in the literature for fraud detection were employed. Data mining methods, including Decision Trees, Artificial Neural Networks, Random Forest Algorithm, and Bayesian Methods, were used to analyze all the data. The analyses revealed that Decision Tree and Random Forest models exhibited the highest accuracy and lowest error rates in detecting fraudulent financial reporting. On the other hand, the Artificial Neural Networks model performed below expectations.