USING APRIORI ASSOCIATION RULES AND DECISION TREE ANALYSIS FOR DETECTING CUSTOMS FRAUD

Abstract

All over the world some people and companies try to avoid taxes whenever possible. Customs are not an exception to this. In this paper we investigate how customs fraud can be detected using data mining on logistics transaction data. We used both the Apriori algorithm for association rules and decision tree analysis to do so. We first transformed the continuous variables using both k-means clustering and CHAID decision tree analysis for the continuous variables in the data set. Analysis of the rules detected by our analysis indicates that it is possible via this methodology to find indicators of customs fraud cases. Moreover, it was possible to describe in detail the situation in which the fraud occurred (product type, country of origin, e-shipper, the weight and the price of the products shipped). The results of the decision tree analysis proved to verify the connection between both Apriori models used and showed similar results, and the evaluation using the classification measures (Accuracy, Precession, Recall and F1) based on the confusion matrix shows high percentages. This confirms the value of our conclusions.