A Boosted Deep ConVNet embedding Long Short Term Memory with Synthetic Minority Oversampling Techniques as Foiling Model for Payment Card Fraud
Journal of International Financial Trends · 2025
DOI: 10.55578/jift.2506.005 ↗Kurz gefasst
Fachartikel in der Zeitschrift Journal of International Financial Trends, 2025. Verfasst von Lateef Salaudeen und 3 weiteren Autor:innen.
Abstract (Originalsprache)
Payment card fraud with contemporaries have persisted as strain challenges bedeviling the financial institutions. With hi-tech thieves exploiting flaws in the digital fraud prevention and detection system to prompts derogatory effects of unquantifiable financial losses, cash back, and customer frictions. An ideal Fraud Detection System (FDS) and countermeasure is require for mitigating these concerns. As a result, several scholars anticipated statistical methods, rule-based approaches with many others for detection. But, majority of these approaches suffers from imbalance data distribution, high dimensionality with sparsity challenges, and real-time detection. This study recommended enhanced Deep ConVNet embedding Long Short-Term Memory and resampling method of SMOTE (DCNN-LSTM+SMOTE) as potential solution. The model is design and implemented on Google Colab platform with GPU; where Tensorflow is used for the DL and Scikit learn for ML models respectively, and Python as the modelled language. Firstly, baseline experiment is steered on two orthodox ML models of Random Forest (RF), Logistics Regression (LR) for feature selection and engineering. While probing kaggle dataset obtained; comprising 284,807 records with 31 field’s features. This dataset is very imbalance with data distribution sort of 0.17% deceitful and 99.8% non-deceitful. Second trialing is conduct; using SMOTE techniques to balance the dataset sort distribution and improved on used LR, RF, with other models such as Isolation forest, Artificial Neural Network (ANN), Multiple Layer Perceptron’s (MLP), Light Gradient Boosting Machine (LGBM), Deep ConVNet and Long Short-Time Memory. In testing efficacy of these models, confusion matrix performance evaluation metrics is delved. This revealed the outcome of the balancing model trial; that described the proposed DCNN-LSTM+SMOTE superclass performance against other models. Where, its accuracy score and prevalence result is 99.8% distinctly. The model offered the second least Error Rate of 0.2%. With 99.9% of recall, True Negative Rate (TNR), Precision and F1-Scores outcomes respectively. Cohen Kappa result is 99.6% and the false positive rate of 0.00%. This result validates the developed model as remarkable in performance, when compared with benchmark studies; and it is promising in the classification of fraudulent credit card transaction in financial institution.
Zitieren
APA
Salaudeen, L., Gabia, D., Garbaa, M., & Surua, H. U. (2025). A Boosted Deep ConVNet embedding Long Short Term Memory with Synthetic Minority Oversampling Techniques as Foiling Model for Payment Card Fraud. Journal of International Financial Trends. https://doi.org/10.55578/jift.2506.005
Vancouver
Salaudeen L, Gabia D, Garbaa M, Surua HU. A Boosted Deep ConVNet embedding Long Short Term Memory with Synthetic Minority Oversampling Techniques as Foiling Model for Payment Card Fraud. Journal of International Financial Trends. 2025. doi:10.55578/jift.2506.005
BibTeX
@article{Salaudeen2025,
author = {Salaudeen, Lateef and Gabia, Danlami and Garbaa, Muhammad and Surua, Hassan Umar},
title = {A Boosted Deep ConVNet embedding Long Short Term Memory with Synthetic Minority Oversampling Techniques as Foiling Model for Payment Card Fraud},
journal = {Journal of International Financial Trends},
year = {2025},
doi = {10.55578/jift.2506.005}
}
EUVIMED