Online Payment Fraud Detection Model Based On SMOTE-Weighted Stacking Framework

Abstract

Online payment fraud detection plays a key role in protecting public property and curbing economic crimes. To solve the problem of sample imbalance in fraud detection, a classification model framework based on composite minority oversampling and weighted stack ensemble is proposed. SMOTE algorithm is used to synthesize and oversample minority samples, which effectively mitigates the impact of data distribution imbalance on classification performance. At the same time, weighted Stacking ensemble strategy is used to fuse the prediction results of multiple base learners, which improves the prediction accuracy of the model and enhances its robustness. It should be noted that both the base model and the meta-model adopt non-parametric modeling methods in this framework, which avoids the potential impact of model default bias on the integration effect. Experimental results based on real transaction data sets show that the proposed model has significant advantages over traditional ensemble learning methods in precision, recall and F1 - score under different sample imbalance ratios.