Authors:
R. Sujeetha, V. Sahaya Sakila

Addresses:
Department of Computer Science and Engineering, SRM Institute of Science and Technology, Ramapuram, Chennai, Tamil Nadu, India.

Abstract:

Credit-card fraud continues to be an ongoing and ever-changing threat to the financial sector, further fueled by the increase in online transactions and ever-changing fraud schemes. Traditional rule-based or individual machine learning approaches may not be effective at detecting fraud due to continually evolving patterns and imbalanced datasets, where legitimate transactions greatly outnumber fraudulent ones. To overcome the challenges mentioned above, this research proposes the X-RANSM, a hybrid stacking ensemble framework to improve fraud detection accuracy and robustness. Using SMOTE (Synthetic Minority Over-sampling Technique) to address class imbalance and merging the base learners with a meta-classifier to maximise final predictions, the proposed framework combines a stacked architecture of Random Forest, XGBoost, and an Artificial Neural Network (ANN) as base learners. A publicly available credit card transaction dataset is used to train and test the system, revealing that only 0.172% of observations exhibit fraudulent behaviour. X-RANSM outperforms baseline samples generated by conventional models on evaluation metrics such as Precision, Recall, F1, and AUC-ROC across all sample sizes. The X-RANSM offers a scalable, flexible mechanism for real-world credit card fraud detection that better generalises across the uneven distribution of credit card transactions by leveraging both data augmentation and multi-model learning.

Keywords: Credit Card; Fraud Detection; Imbalanced Data Classification; Stacking Ensemble Learning; Artificial Neural Network; Synthetic Minority; Oversampling Technique; Deep Learning; Machine Learning.

Received on: 24/12/2024, Revised on: 19/03/2025, Accepted on: 09/05/2025, Published on: 09/12/2025

DOI: 10.69888/FTSCS.2025.000525

FMDB Transactions on Sustainable Computing Systems, 2025 Vol. 3 No. 4, Pages: 233-241

• Views : 41
• Downloads : 6

Download PDF