Abstract

As the global economy continues to expand, international trade is growing at an unprecedented rate, resulting in a surge of cargo movements across borders. With so much activity taking place, it's imperative that customs systems operate intelligently to identify potential risks in every transaction and highlight any potential fraud or manipulation that may be occurring. Customs play a crucial role in supporting legitimate trade, protecting society, and promoting sustainable economic development. However, it is unfortunate that some companies take advantage of the facilities provided by customs departments to avoid paying their duties or engaging in smuggling from the free zone to local market activities. According to data provided by the customs departments in the UAE, there has been a concerning increase in the number of such payment defaults over the years. This not only undermines the fairness and integrity of the customs system but also has a negative impact on the broader economy and society at large. Therefore, customs authorities must remain vigilant and take proactive measures to prevent and deter such illicit activities. By doing so, they can ensure that legitimate businesses are not put at a disadvantage and that the benefits of international trade are shared fairly and equitably. CRISP-DM (Cross-Industry Standard Process for Data Mining) is a widely-used framework for conducting data mining and machine learning projects which will be utilized here. This research is comparing the performance of multiple machine learning techniques including Linear Discriminant Analysis LDA, Random Forest (RF), Quadratic Discriminant Analysis (QDA), Logistic Regression (LR), Naive Bayes (NB), Nearest Neighbours Learning Machine (KNN), Support Vector Machines (SVM), Stochastic Adaptive Boosting, Gaussian Processes, and Bagging to examine if we can predict the payment default in the transactional historical data of the free-zone companies. The research proves that machine learning can be used to predict the payment defaults behaviour, and that the Random Forest model consistently performed better than other evaluated models.

Library of Congress Subject Headings

Customs administration--Data processing; Data mining; Machine learning

Publication Date

10-2023

Document Type

Thesis

Student Type

Graduate

Degree Name

Professional Studies (MS)

Department, Program, or Center

Graduate Programs & Research

Advisor

Sanjay Modak

Advisor/Committee Member

Khalil Al Hussaeni

Campus

RIT Dubai

Plan Codes

PROFST-MS

Download

Share

COinS