Improving Fraud Detection Systems Using Deep Learning with Resampling Techniques
Abstract
Credit card fraud detection remains challenging because real transaction data are extremely imbalanced, where fraudulent cases represent only a tiny fraction of total observations. Models trained on such skewed data can achieve very high overall accuracy while still failing to detect fraud reliably, limiting practical usefulness. This study investigates whether resampling-assisted deep learning can improve minority-class (fraud) detection without generating an impractical false-alarm burden. Using the publicly available Kaggle Credit Card Fraud Detection dataset (284,807 transactions with 492 fraud cases), we evaluate three deep learning architectures—Multilayer Perceptron (MLP), Deep Belief Network (DBN), and Convolutional Neural Network (CNN)—under three training settings: no resampling, Random Under-Sampling (RUS), and Synthetic Minority Over-Sampling Technique (SMOTE). The novelty of this work lies in a controlled comparison of SMOTE versus RUS across multiple deep architectures under a consistent preprocessing pipeline, where resampling is applied only to the training set to prevent information leakage and preserve realistic testing. Model performance is assessed using accuracy together with precision, recall, and F1-score to reflect rare-event detection priorities.The results show that RUS increases fraud recall (0.90–0.92) across models but yields very low precision (0.03) and low F1-scores (0.05–0.07), indicating excessive false positives that reduce deployment feasibility. In contrast, SMOTE produces a balanced improvement in fraud detection while maintaining very high accuracy. DBN + SMOTE achieves the best overall balance with 99.89% accuracy, 0.63 precision, 0.86 recall, and the highest F1-score of 0.73, while MLP + SMOTE achieves the highest accuracy (99.90%) with 0.59 precision, 0.86 recall, and F1-score of 0.70; CNN + SMOTE also performs competitively (99.85% accuracy, 0.54 precision, 0.81 recall, F1-score 0.65). These findings demonstrate that SMOTE-assisted deep learning provides a more deployable precision–recall trade-off than RUS for imbalanced fraud detection, improving fraud recognition while controlling false alarms.
Keywords: Fraud detection, Deep learning, SMOTE, Random under-sampling (RUS), Imbalanced classification.
Full Text:
PDFReferences
CHANG V., ALI B., GOLIGHTLY L., GANATRA M. A., and MOHAMED M. Investigating credit card payment fraud with detection methods using advanced machine learning. Information, 2024, 15(8): 478. https://doi.org/10.3390/info15080478
ALBALAWI T., and DARDOURI S. Enhancing credit card fraud detection using traditional and deep learning models with class imbalance mitigation. Frontiers in Artificial Intelligence, 2025, 8: 1643292. https://doi.org/10.3389/frai.2025.1643292
WU Y., WANG L., LI H., and LIU J. A deep learning method of credit card fraud detection based on continuous-coupled neural networks. Mathematics, 2025, 13(5): 819. https://doi.org/10.3390/math13050819
AHMED M., ANSAR K., MUCKLEY C. B., KHAN A., ANJUM A., and TALHA M. A semantic rule based digital fraud detection. PeerJ Computer Science, 2021, 7: e649. https://doi.org/10.7717/peerj-cs.649
BTOUSH E. A. L. M., ZHOU X., GURURAJAN R., CHAN K. C., GENRICH R., and SANKARAN P. A systematic review of literature on credit card cyber fraud detection using machine and deep learning. PeerJ Computer Science, 2023, 9: e1278. https://doi.org/10.7717/peerj-cs.1278
RTAYLI N., and ENNEYA N. Enhanced credit card fraud detection based on SVM-recursive feature elimination and hyper-parameter optimization. Journal of Information Security and Applications, 2020, 55: 102596. https://doi.org/10.1016/j.jisa.2020.102596
CHAWLA N. V., BOWYER K. W., HALL L. O., and KEGELMEYER W. P. SMOTE: Synthetic minority over-sampling technique. Journal of Artificial Intelligence Research, 2002, 16: 321–357. https://doi.org/10.1613/jair.953
HE H., and GARCIA E. A. Learning from imbalanced data. IEEE Transactions on Knowledge and Data Engineering, 2009, 21(9): 1263–1284. https://doi.org/10.1109/TKDE.2008.239
GUO H., LI Y., SHANG J., GU M., HUANG Y., and GONG B. Learning from class-imbalanced data: Review of methods and applications. Expert Systems with Applications, 2017, 73: 220–239. https://doi.org/10.1016/j.eswa.2016.12.035
HAFEZ I. Y., HAFEZ A. Y., SALEH A., ABD EL-MAGEED A. A., and ABOHANY A. A. A systematic review of AI-enhanced techniques in credit card fraud detection. Journal of Big Data, 2025, 12(1): 6. https://doi.org/10.1186/s40537-024-01048-8
ZIOVIRIS G., KOLOMVATSOS K., and STAMOULIS G. Credit card fraud detection using a deep learning multistage model. The Journal of Supercomputing, 2022, 78(12): 14571–14596. https://doi.org/10.1007/s11227-022-04465-9
FU K., CHENG D., TU Y., and ZHANG L. Credit card fraud detection using convolutional neural networks. In: Neural Information Processing (ICONIP 2016), 2016, pp. 483–490. https://doi.org/10.1007/978-3-319-46675-0_53
KUMARI S. Enhanced deep neural network with SMOTE for credit card fraud detection. EAI Endorsed Transactions, 2025. https://doi.org/10.4108/eai.28-4-2025.2357972
DAVIS J., and GOADRICH M. The relationship between Precision–Recall and ROC curves. In: Proceedings of the 23rd International Conference on Machine Learning (ICML), 2006, pp. 233–240. https://doi.org/10.1145/1143844.1143874
DAL POZZOLO A., CAELEN O., JOHNSON R. A., and BONTEMPI G. Calibrating probability with undersampling for unbalanced classification. In: 2015 IEEE Symposium Series on Computational Intelligence (SSCI), 2015, pp. 159–166. https://doi.org/10.1109/SSCI.2015.33
FAWCETT T. An introduction to ROC analysis. Pattern Recognition Letters, 2006, 27(8): 861–874. https://doi.org/10.1016/j.patrec.2005.10.010
SAITO T., and REHMSMEIER M. The precision-recall plot is more informative than the ROC plot when evaluating binary classifiers on imbalanced datasets. PLOS ONE, 2015, 10(3): e0118432. https://doi.org/10.1371/journal.pone.0118432
Refbacks
- There are currently no refbacks.
