Jul 2026· Science Journal of University of Zakho· Vol 14, pp. 494-503· 0 citations· 25 references
TL;DR
The results indicate that the CNN model can capture the hidden patterns of fraud well when compared to traditional methods, and they are less dependent on manual feature engineering and adaptive to evolving fraud strategies.
Abstract
Financial Fraud threatens global financial stability gravely. Fraud comes with important economic losses and has negative effects on public trust. Traditional rule-based systems and older methods of machine learning cannot detect these patterns of fraud, especially when facing complex or continuously changing fraud patterns in the era of high-speed and high-volume transactions. This research proposes a deep learning technique for the detection of financial fraud by using Convolutional Neural Networks on reshaped tabular transaction data. The proposed CNN model utilizes spatial feature extraction by reprocessing 1D financial records into 2D matrices. This method aims in recognizing local relationships among features as indicative of fraudulent activity. Preprocessing steps for this proposed system include class balancing, scaling of features, and one-hot encoding. It is tested on a highly imbalanced credit card fraud dataset from Kaggle. Then, the classification accuracy was 99.23%. An accuracy of 98.6%, a recall of 97.9%, an F1-score of 98.25%, and an AUC-ROC of 0.992 were also reported. These results indicate that the CNN model can capture the hidden patterns of fraud well when compared to traditional methods. Furthermore, they are less dependent on manual feature engineering and adaptive to evolving fraud strategies.
Detection of credit card frauds is a significant activity in the current financial systems with the rates of electronic transactions escalating and the fraudsters upgrading their fraud techniques as they continually develop. The fact that it is also extremely imbalanced with fraudulent transactions making only a small fraction of the overall data only exacerbates the problem. Hence, there is a need to have effective and smart mechanisms of detection that will guarantee the security of transactions and minimize losses incurred due to financial changes. The objective of the paper is to create and test a deep learning-based credit card fraud detector with the help of Artificial Neural Networks and convolutional neural networks architecture. To meet this goal, a real-life credit card transaction dataset was utilized and until a number of data pre-processing methods were employed that comprised; data normalization, dealing with data imbalances with Synthetic Minority Oversampling Technique, and dimensionality reduction with Principal Component Analysis. The practical section of the paper is based on the implementation and the comparison of three deep learning models, namely, Artificial Neural Network, VGG16, and VGG19. The models were trained and evaluated by normal performance measures which were accuracy, precision, recall and F1-score. It has been experimentally demonstrated that Artificial Neural Networks model depicted the greatest detection precision of 99%, more than both the VGG16 and VGG19 models. These findings validate that deep learning techniques can be used to detect fraudulent credit card transactions and deployed in real time systems of fraud detection.
Deepika Tiwari, Meenakshi Nawal, N. Neeraj et al.· Journal of Dynamics and Cont...· 1 citation
Credit card fraud has emerged as one of the most significant challenges faced by financial institutions due to the rapid growth of digital banking and online transactions. Traditional machine learning algorithms often struggle to maintain high detection accuracy while minimizing false positives, particularly in highly imbalanced transaction datasets. This paper proposes an interpretable deep learning framework based on the TabNet architecture for enhancing credit card fraud detection in banking systems. Unlike conventional neural networks, TabNet employs sequential attention mechanisms to perform feature selection dynamically, enabling superior learning from structured tabular transaction data while providing model interpretability. The proposed methodology includes comprehensive data preprocessing, missing value handling, feature scaling, class imbalance correction using SMOTE, and hyperparameter optimization. Experimental evaluation is conducted on the publicly available European Credit Card Fraud Detection dataset. Performance is evaluated using Accuracy, Precision, Recall, F1-score, ROC-AUC, PR-AUC, and Matthews Correlation Coefficient (MCC). Experimental results demonstrate that TabNet outperforms traditional neural networks and popular machine learning algorithms by achieving high fraud detection accuracy with significantly reduced false alarms while maintaining excellent interpretability suitable for financial regulatory requirements. Recent studies have similarly reported strong performance of TabNet for banking fraud detection, highlighting its suitability for structured financial datasets and explainable AI applications.
G Srividhya, Dr. S Siva Sankara Rao· International Journal of Eng...· 0 citations
Detecting credit card fraud remains a critical issue in financial security, demanding models that excel on imbalanced tabular transaction data while balancing accuracy, interpretability,
and efficiency.
This paper proposes a novel hybrid CNN-AdaBoost framework, where convolutional neural networks extract intricate, complex patterns from raw transactional features, and AdaBoost
performs robust iterative classification to refine predictions.
We evaluate the approach on standard credit card fraud benchmarks, achieving an overall accuracy of 95.43% alongside a strong F1-score of 92.75%, surpassing baselines methods. Ablation studies also conducted to confirm synergy: standalone CNN yields 93.12% accuracy, while pure AdaBoost reaches 93.58%, highlighting how feature fusion addresses individual shortcomings in handling noise and class imbalance.
The novelty stems from repurposing CNN’s convolution layers to capture local, complex patterns directly from raw tabular transaction data, feeding these rich representations into AdaBoost for adaptive error correction. This hybrid uniquely blends deep feature learning’s nuance with ensemble boosting’s efficiency and interpretability, outperforming standalone methods on imbalanced fraud detection. Prior works rarely fuse these for tabular domains, yielding our 95.43% accuracy benchmark.
This provides valuable insights into the design of hybrid models for structured and tabular data classification tasks and establishes a strong benchmark for future research in this domain.
The rise in digital transactions, the volume of transactions has increased leading to rapid growth in the fraud detection rate in financial networks. Challenges in fraud detection include imbalanced data, high transaction feature dimensionality, evolving fraudulent patterns and the need for real-time decision making. However, traditional fraud detection approaches fall short in areas of imbalanced datasets handling, feature selection, detection accuracy and high false negative rate. In this study, an AI-driven adaptive fraud detection system is proposed for financial networks using European Cardholder Dataset. The proposed framework applies data preprocessing techniques including data cleaning, data normalization and Synthetic Minority Over-sampling Technique (SMOTE) to address the dataset imbalance issue. We apply Principal Component Analysis (PCA) for feature extraction to convert transaction data to meaningful features. Filter-based feature selection methods such as Correlation and Chi-square are applied to identify fraudulent transaction data features with the highest dependency and correlation. Random forest and SVM are implemented as classifiers for fraudulent transactions. Random forest utilizes ensemble learning to strengthen the robustness of the detection system. SVM increases classification accuracy for identifying samples with complicated distribution patterns. Accuracy, precision, recall, F1-score and receiver operating characteristic area under the curve (ROC-AUC) are used to evaluate the system performance. The system provides a high accuracy with a very low false negative rate, which makes it viable for real time fraud detection in financial networks.
G. M, S. S, Sarulatha G et al.· 2026 6th International Confe...· 0 citations
Financial fraud classification cases such as credit card fraud and bitcoin fraud have highly imbalanced data problems that the oversampling data of fraud class is necessary. Financial transactions could have different attributes. In a credit card transaction, the attributes could represent a nominal amount, transaction period infor- mation, the status of deposits or other types like withdrawals or refunds, and more detailed information. In the financial transaction of bitcoin, the attributes could rep- resent the number of nodes, transaction fee, output volume, and aggregated figures. The various characteristics of attributes in financial fraud data require an adaptable oversampling method so that the classification model can perform well. An Ensemble Oversampling method is proposed as a general context approach to handling finan- cial fraud classification in credit cards and bitcoin. The proposed method combines generative with traditional approaches such as GAN, SMOTE, and ADASYN. In the classification step, Deep Learning algorithms such as CNN and LSTM are applied to provide better performance. The genetic algorithm is used to optimize Deep Learn- ing hyperparameters. The evaluation was carried out by comparing four scenarios, i.e., without oversampling, using oversampling with GAN, SMOTE, ADASYN, orig- inal data, and Ensemble Oversampling. The combined oversampling of GAN and SMOTE with the CNN classifier model produces the highest evaluation score of all scenarios with an average F1-Score value of 0.995 and Kappa Statistics of 0.990. It shows that augmented data quality does affect prediction performance, and Ensem- ble Oversampling technique could be considered to improve classifier performance in financial fraud data.
Moch Deny Pratama, Agus Budi Raharjo, Diana Purwitasari· IPTEK: The Journal for Techn...· 0 citations
The rapid growth of digital financial services has significantly increased the volume of online transactions, making fraud detection a critical challenge for financial institutions. Traditional machine learning models often provide strong predictive performance but lack interpretability, limiting trust and practical adoption in financial decision-making. This study proposes an Explainable Artificial Intelligence (XAI)-based fraud detection framework for FinTech transactions using the Kaggle Credit Card Fraud Detection dataset containing 284,807 transactions, including 492 fraudulent cases. To address severe class imbalance, Synthetic Minority Oversampling Technique (SMOTE) was applied, increasing the dataset to 568,630 balanced instances. Data preprocessing involved feature scaling and train–test splitting prior to model training. Three machine learning algorithms—Logistic Regression, Random Forest, and Extreme Gradient Boosting (XGBoost) were developed and evaluated using accuracy, precision, recall, F1-score, and ROC-AUC metrics.
The experimental results demonstrate strong predictive performance across all models. Logistic Regression achieved 94.50% accuracy, 97.32% precision, 91.51% recall, 94.33% F1-score, and a ROC-AUC of 94.50%. Random Forest produced the highest overall performance with 99.99% accuracy, 99.98% precision, 100.00% recall, 99.99% F1-score, and 99.99% ROC-AUC. XGBoost also achieved excellent results with 99.97% accuracy, 99.94% precision, 100.00% recall, 99.97% F1-score, and 99.97% ROC-AUC. To improve model transparency, SHapley Additive exPlanations (SHAP) and Local Interpretable Model-Agnostic Explanations (LIME) were integrated with the XGBoost model to provide both global and local interpretability. SHAP analysis identified transaction amount and several transformed principal component features as the most influential predictors of fraudulent behavior, while LIME provided instance-level explanations for individual fraud predictions. Feature importance analysis from Random Forest and XGBoost further validated the consistency of the most influential variables.
The findings demonstrate that combining high-performing machine learning models with explainable AI techniques can significantly enhance fraud detection accuracy while maintaining transparency and interpretability. The proposed framework offers a reliable and practical approach for intelligent fraud prevention in financial technology systems and supports trustworthy decision-making in real-world financial environments.
D. Aina, J. A. Ayeni, A. Ogunjobi et al.· International journal of res...· 0 citations