A Novel Unsupervised Classification Method for Customs Fraud Detection

Abstract

Objectives: In this paper, a very light and straightforward algorithm is proposed for customs fraud detection. Methods/Analysis: in order to fraud detection we have proposed our algorithm based on unsupervised methods. Our approach is a combination of data clustering methods, Mahalanobis distance classifier, K Nearest Neighbor (KNN) method, and density-based methods. Findings: The results showed that the proposed method was able to accurately identify frauds, as more than 73 percent of high-risk goods that the proposed method is detected, has been violated. It is faster and more rapid than the other methods. The method requires less processing than other methods, and more than 30 percent CPU usage has been improved. The approach is independent of distribution and scattering of data samples. It also has the ability to work with samples by different clusters, densities, and no limitation on dimension of data. Novelty of the Study: For the first time, an unsupervised method is used for finding the frauds in customs. Application/Improvements: One of the most important applications of the results of this study is the Customs Risk Management System. Also, the proposed approach will enhance the ability of fraud detection in trade.