Fraud Detection in Audit Data Using Machine Learning Algorithm

Accounting records are regarded as analytical documents that financial businesses produce on a regular basis to provide multiple viewpoints on their financial condition. Since many investors and auditors rely heavily on these reports to make decisions, certain institutions may falsify them in order to deceive the public and perpetrate fraud. The principal goals of discovery of accounting records forgery are to recognize anomalies brought about by these biases and separate reports that are likely to include fraud from those that are not. Among the most common approaches for analyzing data in this field is binary classification, however it requires an established tagged datasets, which is sometimes unobtainable in real-world situations because fraudulent samples are uncommon. This study presents a novel strategy that addresses the great degree of scale in the parameter area and resolves the shortage of non-fraudulent samples. It is based on machine learning and ensemble models. Additionally, a new datasets is created by gathering ten Iranian bank's annual financial statements and then extracting the three kinds of attributes this study suggests.