Adopting Global Machine Learning Models for Credit Card Fraud Detection in Libyan Banks: Challenges and Strategies

Abstract

This research investigates how global machine learning (ML) models can be adopted to detect credit card fraud in the Libyan banking sector, which currently lacks advanced fraud detection systems and AI policy frameworks. The researchers reviewed international and regional applications of ML for fraud detection and identified a significant gap in Libya. Using a public European credit card dataset, the researchers trained and evaluated four ML models: XGBoost, Gradient Boosting, Neural Networks, and Isolation Forest. Among them, XGBoost achieved the highest accuracy (99.98%), recall (100%), precision (99.96%), and F1-score (99.98%). The Neural Network followed with strong accuracy (99.97%) and recall (99.99%), precision (99.95%), and F1-score (99.97%). Gradient Boosting and Isolation Forest showed high accuracy but low recall, indicating limited effectiveness in detecting positive fraud cases. Based on these results, the researchers recommend XGBoost as the most suitable model for initial deployment in Libyan banks and emphasizes the need for future adoption using localized transaction data.