FRAUD DETECTION USING DATA ANALYTICS: A CASE STUDY OF UNDER INVOICING IMPORTATION FRAUD IN INDONESIA

Abstract

Fraud detection is a big concern for all the government agencies. In customs areas, fraud detection is needed to ensure that there is no leakage in state revenue, one of which is caused by the under invoicing importation fraud. The data analytic implementations have been used in many studies to handle problems in big data and give solutions. This study aims to explain how data analytics can be implemented to detect the under invoicing importation fraud. Several variables were included in this study, including the variables that show the risk level of importers, commodities, suppliers, and the exporter countries. This study compared various machine learning models including Logistic Regression, Decision Trees, Random Forest, Extreme Gradient Boost, Artificial Neural Networks, Gaussian NB, and K-nearest Neighbors. To evaluate the models, this study measures the performance of the models by comparing accuracy score, precision score and log loss score. The result shows that the Xtreme Gradient Boost performs best in detecting under invoicing fraud with accuracy score at 63%, precision score at 63% and log loss score at 62%. As far as we know, this has been the first work to compare a number of machine learning models to create under invoicing fraud detection. The results of this study will assist examiners in the import clearance process by providing an early warning of the under-invoicing transaction. It can lead to more effective and efficient examination, so that customs agencies can perform well in their service and inspection functions, despite the limited resources