Table 1 List of related works.
From: RABEM: risk-adaptive Bayesian ensemble model for fraud detection
References | Dataset | Methodology | Limitations | Results |
|---|---|---|---|---|
Financial datasets for fraud detection | Blockchain, smart contracts, and incremental ML model updating | Privacy concerns, mining difficulty, and data volume sensitivity | 98.93% accuracy, 98.22% F-beta score, efficient mining | |
Three distinct datasets with online payment transactions | Compare twelve machine learning algorithms for fraud detection | Limited algorithm scalability, dataset diversity not specified | Gradient boosting consistently achieved an accuracy of 96.8% | |
6.3 million unbalanced financial transactions, five types | Stacked logistic regression, SVM, random forest, and sampling strategies | Data imbalance, missing entries, and limited real-world validation | 97% recall, 87% accuracy on synthetic data | |
Credit card transaction data with fraud labels | Weighted average ensemble combining multiple classifier models | Potential overfitting and real-world applicability are not discussed | Accuracy of 99%. Meanwhile, other classifier ensembles achieve accuracies of 97% and 98% | |
Imbalanced credit card transaction data with fraud labels | XGBoost with data balancing techniques, such as oversampling | Imbalanced data can significantly impact model performance if not properly balanced | XGBoost with oversampling achieved a precision and accuracy of 99% | |
Real-world credit card transaction data for fraud | Bayesian-optimized Extremely Randomized Trees using TPE | Potential overfitting with complex, imbalanced transaction data | TP-ERT achieved a precision of 0.97 and an F1-score of 0.95 | |
Analyzing imbalanced credit card transactions for fraud detection | Random Forest with SMOTE and entropy criterion | Trade-offs in false positives lack real-world testing | Precision: 0.95, recall: 0.79, F1-score: 0.86 | |
Kaggle and IEEE-CIS credit card fraud datasets | Unsupervised Anomaly Detection Using Attention and Generative Adversarial Networks (GANs) | Lower recall indicates challenges in detecting all types of fraud | Precision: 0.9795, recall: 0.7553, F1-score: 0.852 | |
The Bank Account Fraud (BAF) transaction dataset was used | Advanced ensembling techniques, especially stacking models, were applied | Precision-recall trade-offs lack real-time deployment validation | The stacking model achieved 0.98 accuracy in detecting fraud | |
Credit Card Fraud Detection Dataset with Class Imbalance | Credit Card Fraud Detection Dataset with Class Imbalance | Credit Card Fraud Detection Dataset with Class Imbalance | Accuracy: 0.97, AUC: 0.92, F1-score: 0.11 | |
Imbalanced credit card transaction data for fraud detection | Soft voting ensemble with sampling techniques for classification | Class imbalance persists, though misclassification costs are reduced | Precision: 0.987, recall: 0.969, F1-score: 0.8764 |
