Original scientific paper

https://doi.org/10.17559/TV-20241013002057

Navigating the Fraud Frontier: Machine Learning Solutions for Credit Card Security

Hye Jin Kim ; Dept. of Smart Convergence Security, Busan University of Foreign Studies, 65, Geumsaem-ro 485beon-gil, Geumjeong-gu, Busan, Republic of Korea
Jung Soo Rhee ; Dept. of Smart Convergence Security, Busan University of Foreign Studies, 65, Geumsaem-ro 485beon-gil, Geumjeong-gu, Busan, Republic of Korea *

* Corresponding author.

Abstract

Credit cards are essential financial tools that provide convenience and flexibility to both individuals and companies, allowing safe transactions and enabling worldwide trade. However, as the number of transactions increases, the probability of fraud also rises. This presents substantial risks to financial security. It undermines customer confidence. The rapid increase in credit card transactions highlights the urgent need for advanced fraud detection systems. This study examined the efficacy of six prominent machine learning algorithms: Logistic Regression (LR), Random Forest (RF), K-Nearest Neighbours (KNN), Decision Tree (DT), Support Vector Machine (SVM), and Naive Bayes (NB) in detecting fake credit card transactions using a dataset of 568,630 transactions with 30 features collected from European cardholders in the year 2023. The data was gathered from multiple financial institutions and payment processors throughout Europe. The dataset features V1 to V28, which are anonymized transaction details like time and location. The Amount column shows the transaction value, while the Class column indicates if a transaction is fraudulent (1) or not (0). To address the inherent disparity between fake and real transactions, the research utilises thorough preprocessing approaches such as feature scaling and class balancing to ensure the model's strong performance. Thorough data analysis and visual representation were used to find significant transaction patterns and connections among various characteristics. Each algorithm is subjected to thorough training and optimization via the modification of hyperparameters and cross-validation. The evaluation of performance is carried out by analysing metrics such as accuracy, precision, recall, F1 score, and Receiver Operating Characteristic-Area under Curve (ROC-AUC). Results demonstrated that the RF model achieved the highest accuracy of 99.98% and a perfect AUC of 1.000. LR also performed well with an accuracy of 96.56% and an AUC of 0.9935. These results provide vital insights into the practical uses of these algorithms in increasing fraud detection systems and improving financial security, particularly as credit card usage continues to grow.

Keywords

class balancing; classification matrices; financial tools; fraud detection; machine learning

Hrčak ID:

328648

URI

https://hrcak.srce.hr/328648

Publication date:

27.2.2025.

Visits: 796 *