| Abstract [eng] |
Value Added Tax (VAT) is an important source of income in most European countries. In Lithuania, VAT accounts for 44 \% of all income, and that makes the collection of VAT an important task as it is essential to support state allocations and sustainability. However, there are many ways the VAT taxation system can be exploited. In Lithuania, registered VAT taxpayers are obliged to submit monthly VAT declaration form FR0600. Obligation to submit VAT declarations makes such acquired data a perfect choice for the identification of fraudulent activities. The aim of this thesis is to identify possible VAT fraud cases by applying selected unsupervised anomaly detection methods. In order to reliably identify VAT fraud cases, several tasks need to be addressed, including analyzing, implementing, comparing, and summarizing the fraud detection alternatives that exist in the field of unsupervised learning. Fraud is considered to be rare and markedly different from legitimate observations event; therefore, anomaly detection techniques were considered. |
