From chaos to clarity: unraveling credit card fraud with BGVOA-LS

Résumé

Abstract Credit card fraud (CCF) continues to cause significant financial losses worldwide, costs institutions and consumers billions of dollars each year and erodes confidence in digital payments. Fraudulent activities are rare, highly adaptive, and increasingly complex, making them difficult to detect with traditional rule-based or static models. This study presents the binary Griffon vulture optimization algorithm with local search (BGVOA-LS) framework, which integrates optimized feature selection (FS) with machine learning (ML) to improve accuracy and efficiency in fraud identification by selecting the most relevant transaction features and removing noise and redundancy so that classifiers focus on patterns most indicative of fraudulent behavior. To address the severe imbalance between legitimate and fraudulent records, the method applies random under sampling (RUS), which helps reduce the number of false positives and missed detections. The selected features are evaluated via three classifiers across the European, Australian, and PaySim datasets: decision tree (DT), k-nearest neighbors (KNN), and extreme gradient boosting tree (Xgb-tree). BGVOA-LS achieves up to 99.8% accuracy, reduces the feature count by 67%, and outperforms other metaheuristic techniques (MHTs), enabling faster screening, lower computational costs, and more dependable fraud prevention.

Citer ce document

Accès au document

Texte intégral en lecture en ligne, réservé aux abonnés SPHAERO et aux membres de l'institution. Se connecter

Auteur(s)

Statistiques

Consultations : 1

Téléchargements : 0