Abstract

Aim- The study aimed for creating a machine learning algorithm which succeeded customs control procedures through estimating shipments being either illicit or non-illicit. The study aimed at boosting the level of accuracy in determining the critical shipments, which would in turn, increase customs productivity, minimize false positive and false negative, as well as augment security and ensure the smoothness of trade flow, amid a palpable surge in importation rates. Methods- The study implemented the CRISP-DM methodology (Cross-Industry Standard Process for Data Mining), which has a well-structured approach of achieving standard data mining solutions. At the beginning, the efforts have been mainly made on defining both the project and operation targets, resolving the specific issues and developing the control and resources within the customs context. Then, data collection and understanding were the next tasks that the researcher needed to deal with, which held attributes like the number, types, quality, and association. Then data cleaning, transformation, and features construction were involved, which included data preparation for modeling. Modelling phase would be in strongly connected with the ETL process through construction, examination and assessment of models making sure that the model that is selected matches the requirements involving illicit and non-illicit shipments most of all. As the last step, the IBM SPSS Software has been utilized for statistical evaluations, considering ROC curve, precision, and confusion matrix as evaluation metrics for identifying the best suited model in terms of accuracy for predicting a complex dataset of customs operations. Findings- The study revealed that customs authorities were confronted with insurmountable problems particularly the ability to differentiate between safe goods and contraband items, which is getting more challenging while keeping other objects flowing. In this context, the call to arms involved two tasks; the first one was aimed to enhance existing detection processes and the second one was to find out how to apply a neural network model that could specifically address the issues and improve the efficiency of air cargo shipments. The research results showed that the model was effective in detecting illicit shipments that are hidden within the huge range of international air shipments and therefore, it represents a distinctive solution to elevate the security level and defend the economic interests. The developed neural network model acted as a crucial tool in terms of countering the risks related to the national security threats and illicit activities which could have consequentially impacted the trade flows and solidified economic instability. Through correct identification of shipments with a high likelihood of illicit cargo entering customs checkpoints, the model succeeded in reducing the chances of undiscovered contraband successfully passing through, which helped to prevent security breaches and losses. Such capability was especially significant in the cases of high-level customs visits where the repercussions of undetected clandestine shipments include breach of state security and the adverse economic implications. Key performance metrics of the neural network model was its high dependability in shipping contraband packages through its high recall performance measures. This metric had a center place in optimization of customs inspection procedures, which makes a possibility of accomplishment a more targeted and efficient control of entering shipments to a country and minimization of disruption of the legal trade. In addition, harnessing sophisticated technologies such as neural network models help customs agencies to increase their resilience as well as the ability to cope with evolving risks and new smuggling tactics. As a result, they are better placed to increase security and efficiency in the context of constant global trade dynamics advancement.

Library of Congress Subject Headings

Customs inspection--Data processing; Smuggling--Prevention; Neural networks (Computer science); Decision trees; Data mining

Publication Date

5-6-2024

Document Type

Thesis

Student Type

Graduate

Degree Name

Professional Studies (MS)

Department, Program, or Center

Graduate Programs & Research

Advisor

Sanjay Modak

Advisor/Committee Member

Hammou Messatfa

Campus

RIT Dubai

Plan Codes

PROFST-MS

Download

Share

COinS