Aug 2026· Engineering Reports· Vol 8· 0 citations· 22 references
TL;DR
Three deep tabular models, namely, an advanced multilayer perceptron (AdvancedMLP), an attention‐based residual network (AttentionFraudNet), and an advanced residual network (AdvancedResNet), are compared against three traditional machine learning baselines, including Random Forest, Gradient Boosting, and Logistic Regression.
Abstract
Financial fraud detection presents a significant challenge due to the extreme class imbalance inherent in transaction datasets. This paper presents a comprehensive evaluation of deep learning architectures for fraud detection in structured transaction data, with a focus on model performance, interpretability, and operational relevance. Specifically, we compare three deep tabular models, namely, an advanced multilayer perceptron (AdvancedMLP), an attention‐based residual network (AttentionFraudNet), and an advanced residual network (AdvancedResNet), against three traditional machine learning baselines, including Random Forest, Gradient Boosting, and Logistic Regression. The experiments are conducted on the BankSim Simulator dataset, comprising 594,643 transactions with a fraud rate of approximately 1.21%. We evaluate models using multiple metrics, including PR‐AUC, ROC‐AUC, recall at low false positive rates, and precision at various operating thresholds. Our results show that AdvancedMLP achieves the highest PR‐AUC (0.8921 ± 0.0020), while Gradient Boosting attains the best ROC‐AUC (0.9987 ± 0.0001) and the highest recall at 1% FPR (0.9498 ± 0.0087). AttentionFraudNet provides interpretable attention weights that highlight transaction‐level features contributing to fraud predictions, enabling fraud analysts to understand and validate model decisions. Business impact analysis reveals that the proposed models can reduce fraud losses while maintaining false positive rates below 1%, translating to estimated annual savings of approximately 49.2 M$ for a mid‐sized financial institution. All models achieve inference times below 1 ms per transaction on an NVIDIA RTX 3090 GPU, demonstrating computational feasibility for near‐real‐time screening. Our findings highlight the effectiveness of deep tabular models for fraud detection and provide practical guidance for selecting models based on operational requirements and performance trade‐offs.
The study introduces a Stacked Logistic Regression ensemble to combine the predictive capacity of optimized Random Forest and XGBoost base classifiers and reveals that the proposed stacked model performance surpasses both individual base models.
Uduh Israel Akakoh, G. N. Edegbe· FUDMA Journal of Sciences· 0 citations
Insurance fraud detection poses a key challenge due to substantial class imbalance, heterogeneous claim types, and the continual evolution of fraudulent practices. Despite several Machine Learning (ML) approaches having been developed, comparative assessments that simultaneously address predictive performance, calibration stability, cost-effectiveness, and comprehensive statistical analyses remain scarce. To bridge the gap, this study presents a robust evaluation framework for identifying auto insurance fraud that incorporates boosting-based classifiers, advanced deep tabular architectures, and adaptive resampling methods. Six classification models -namely Categorical Boosting (CatBoost), Light Gradient Boosting Machine (LightGBM), eXtreme Gradient Boosting (XGBoost), Attentive Interpretable Tabular Learning Architecture (TabNet), Feature Tokenizer Transformer (FT-Transformer), and Multi-Layer Perceptron Residual Network (MLP-ResNet)- were tested under three data balancing methods (Original, SMOTE, and ADASYN). Contrary to previous studies that focus exclusively on classification performance, the proposed framework includes sensitivity analysis, probabilistic calibration scoring, effect-size evaluation, expected-cost analysis, Pareto frontier optimization, and statistical significance tests. The comparative evaluation reveals that no single model systematically outperforms all others. CatBoost with ADASYN provides the most balanced predictive performance, while TabNet presents superior probabilistic prediction quality and the lowest expected cost. These findings show that the optimal model depends on the target objective, whether predictive accuracy or cost-sensitive fraud detection, a result further validated by the Friedman and Nemenyi statistical tests. In addition, the Pareto frontier analysis identifies CatBoost and TabNet as supplementary optimal trade-offs between predictive performance and operational cost. Statistical analysis confirms significant differences among the evaluated configurations, attesting to the robustness of the findings. Overall, the proposed framework highlights the complementary strengths of boosting-based approaches and TabNet, offering a reliable and cost-effective method for insurance fraud detection, depending on the targeted operational objective.
Click fraud remains one of the most pressing threats in digital advertising, inflating campaign costs and distorting marketing analytics. Conventional detection approaches are frequently unable to adapt to sophisticated and evolving fraud patterns. This study proposes an explainable hybrid framework that integrates deep learning and gradient boosting for ad click fraud detection using the Kaggle Ad Click Fraud Detection Dataset (5,000 records, 21 features). During preprocessing, multicollinearity was systematically removed through Pearson correlation analysis (threshold ρ > 0.85), reducing the feature set from 21 to 14 highly discriminative attributes. Random Under-Sampling (RUS) and SMOTE were applied to correct class imbalance. Thirteen classical machine learning and deep learning models were benchmarked, including CNN, DNN, RNN, LSTM, GRU, and hybrid LSTM–GRU networks. A Voting Classifier combining XGBoost and Bagging with Decision Tree served as the proposed architecture, achieving 100% accuracy, precision, recall, and F1-score. Explainability was incorporated through LIME (local, per-prediction waterfall explanations) and SHAP (global feature importance rankings). A Flask-based web interface enables real-time fraud prediction. The architecture demonstrates inference latency below 50 ms per request, making it suitable for large-scale, real-time advertising platforms.
S. S, J. R· 2026 7th International Confe...· 0 citations
Abstract The rise in online transactions has made credit card fraud a significant global concern, necessitating detection strategies that are both highly accurate and practically viable. While existing literature extensively explores machine learning techniques to address class imbalance, most studies optimize for traditional statistical metrics, overlooking the asymmetric financial costs and strict operational constraints of real-world fraud detection. This study bridges this gap by proposing a comprehensive, cost-sensitive ensemble framework evaluated on a real-world European cardholder dataset. We move beyond the traditional F 1 score by adopting the cost-sensitive F β metric to reflect real financial impact. Through exhaustive benchmarking, we show that while eXtreme Gradient Boosting (XGBoost) combined with Borderline SMOTE achieves the highest single-model performance, our proposed soft-voting ensemble integrating Logistic Regression and Random Forest with SMOTE delivers the best overall performance (F β = 0.8287). To ensure practical viability, we introduce a Top-K operational constraint evaluation reflecting limited human investigation bandwidths. Additionally, an ablation study demonstrates that there is no universal remedy for class imbalance; optimal interventions are highly model-dependent. Finally, by validating our framework on a feature-transparent simulated dataset, model interpretability analysis reveals the ensemble’s capacity to capture the critical importance of environmental risk factors, shifting the focus beyond solely customer-centric anomalies.
Xin-Yue Fan, T. Boonen· Asia-Pacific Journal of Risk...· 0 citations
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.
G. Srividhya, S. S. Rao· International Journal of Eng...· 0 citations
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, outperforming standalone methods on imbalanced fraud detection.
Santosh Nirmal, Poonam L Rakibe, Pramod D. Patil· Electronics ETF· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
