Validation-Stage Combinatorial Fusion Analysis for Imbalanced Credit-Card Fraud Detection
Researchers have explored Combinatorial Fusion Analysis (CFA) to improve credit-card fraud detection, particularly for imbalanced datasets. Their study on the IEEE-CIS Fraud Detection benchmark found that CFA, by selecting and weighting a subset of diverse classifiers like Random Forest, XGBoost, and LightGBM, can enhance performance metrics such as AUC-ROC, AUPRC, and F1 score. The analysis also indicated that synthetic data augmentation using CTGAN did not improve results and could potentially degrade model performance. AI
IMPACT Introduces a novel fusion technique that could improve the accuracy of fraud detection systems by better handling imbalanced datasets.