Document Type
Article
Abstract
Credit card fraud detection is a critical task in financial systems, especially given the rarity and evolving nature of the fraudulent behavior. The highly imbalanced class levels of the fraudulent and non-fraudulent transactions make it a challenging classification problem to solve. This study investigates the effectiveness of machine learning models: Logistic Regression, XGBoost, and Multi-Layer Perceptron (Neural Network), evaluated under temporal retraining and fine-tuning scenarios using a publicly available, highly imbalanced dataset of European credit card transactions. The dataset includes 284,807 transactions, of which only 492 (0.172%) are labeled as fraudulent, making it a well-known example of an imbalanced classification problem. The models are evaluated using metrics including precision-recall AUC, ROC-AUC, and class-specific precision and re-call—emphasizing capturing minority fraud cases. The research indicated that temporal retraining preserves or improves recall over time, while fine-tuning enables neural networks to adapt incrementally without the requirement of full training. Among the three models, XGBoost demonstrates the best balance of precision and recall across all setups. These findings highlight the value of continual learning strategies in real-world fraud detection pipelines.
Disciplines
Artificial Intelligence and Robotics | Computer Sciences | Data Science
Publication Date
2025
Language
English
License
This work is licensed under a Creative Commons Attribution-Share Alike 4.0 International License.
Recommended Citation
Khadka, Anamol, "Credit Card Fraud Detection via Model Retraining and Fine-Tuning" (2025). Computer Science and Engineering Student Research. 1.
https://mavmatrix.uta.edu/cse_studentresearch/1