Multi-label Classification Approach to Identify Fraudulence in Financial Statements

Financial statement fraudulent activity is more common as a result of the worldwide economic instability caused by the COVID-19 pandemic, endangering the effective operation of equity marketplaces. Despite the fact that business audit procedures have improved over time, there are more cases of company fraud, which makes it necessary to look into potential early alerts and create efficient platforms for spotting financial fraud. The goal of financial fraud, which is frequently carried out via asset transmissions and presentation, is to obtain credits and lower taxation. The standard of the data is crucial for automatic data assessment systems. Among the particulars that must be considered is the potential for the same data part to be assigned to many classes. Data mining approaches employ business management characteristics and financial data as sources for developing systems that can spot trends or anomalies in a business's financial statements. To accomplish this, well-known AI methods are used in a new multi-label classification setup that takes into account the auditors' remarks in addition to identifying fraudulent instances. For sample proportions of 8:3, 7:4, and 6:5, the outcomes show classification accuracy of 91.8%, 90.4%, and 91.0%, respectively. 90.86% is the precision, 90.78% is the recall, and 90.82% is the F1-value. While recognizing classification just took 0.03 ms, training took 95.82 ms. The findings show that, in comparison to binary classification methods, the suggested multi-label plan yields better outcomes and avoids ambiguous conclusions about the presence of various financial statement modification techniques.

Перевод пока недоступен