Towards Electronic Fund Transfer Fraud Detection Using Machine Learning

Simple item page
AuthorIsmaeel, Hussain
metadata.dc.contributor.departmentUNIVERSITY OF BAHRAIN, College of Information Technology, COMPUTER SCIENCE Department
metadata.dc.date.accessioned2026-09-20T15:48:33Z
Publication date2023
Degree levelMaster
AbstractABSTRACT: Electronic payment methods have significantly evolved and became dominant in the contemporary financial system, offering a variety of seamless payment channels of elec-tronic transactions. The number of electronic payments transactions has taken a drastic leap in the last decade, processing millions of transactions on a daily basis. However, the dramatic increase in digital payment transactions has been accompanied with dreadful number of fraudulent activities which result in substantial financial losses to financial in-stitutions and their consumers. Further, the rapid technological advancement plays a key role in the increase of fraud transactions and are projected to ever grow which posture vital risk. Thus, fraud prevention and detection has become a necessary requirement to ensure reliability and credibility of financial institutions. Data mining and machine learning tech-niques have been prominent strategies by researchers to develop methods to detect frauds and reduce the risk. In that context, designing efficient and practical algorithms have en-countered several challenges including the lack of interpretability of fraud definition, the unavailability of fraud transactional data due to privacy regulation, and the sophisticated indistinguishable characteristics of fraud transactions. In this thesis, we propose a machine learning model to detect fraud Electronic Funds Transfer (EFT) transactions. To assess the model's efficacy, experiments are carried out on a vast simulated transactions dataset span-ning two years which incorporated Synthetic Minority Oversampling Technique (SMOTE) resampling technique along with cross validation using stratified K-fold. The evaluation involved the utilization of multiple metrics to assess the model comprehensively, which comprised the confusion matrix, Receiver Operating Characteristic Area Under the Curve (ROC AUC), Precision-Recall Area Under the Curve (PR AUC), precision, recall, F1-score, and time efficiency. The performance results show employing SMOTE technique improve the model detection rate, and the proposed XGBoost model proved to be the most efficient and effective, achieving a perfect ROC AUC and a PR AUC of 0.994.
metadata.dc.identifier.otherDrupal node 3722
metadata.dc.identifier.otherhttps://digitalrepository.uob.edu.bh/en/dar/towards-electronic-fund-transfer-fraud-detection-using-machine-learning
metadata.dc.identifier.urihttp://165.227.56.122/handle/123456789/6760
Languageen
Linkhttps://digitalrepository.uob.edu.bh/en/_flysystem/fedora/2025-07/00799_towards_electronic_fund_transfer_fraud_detection_using_machine_learning.pdf
TitleTowards Electronic Fund Transfer Fraud Detection Using Machine Learning
TypeMaster's Thesis
CollegeCollege of Information Technology
DepartmentComputer Science

Files

Original bundle

Now showing 1 - 1 of 1
Thumbnail Image
Name:
node-3722-1-00799_towards_electronic_fund_transfer_fraud_detection_using_machine_learning.pdf
Size:
35.2 MB
Format:
Adobe Portable Document Format
Download

Collections

Master's Theses