Enhancing Strategic Decision-Making in Fraud Management: A Dual-Channel Framework with TOPSIS for Credit Card Fraud Detection

Mobile payments and e-commerce, along with credit cards, have become deeply embedded in people’s daily business and economic activities. However, the increase in transaction volume has also led to a rise in fraudulent activities. To address challenges such as extremely low fraud sample ratios, highly imbalanced data distributions, significant noise interference, and strong sequence dependencies, this paper proposes a multi-module collaborative framework for credit card fraud detection. This framework includes four core components: anomaly-aware cleaning, structure-preserving resampling, multi-indicator feature governance, and dual-channel modelling. A multi-attribute decision method based on TOPSIS is introduced for feature selection. Additionally, a dual-channel detection model is developed, comprising a static channel (random forest) and a dynamic channel (LSTM), to model non-sequential and sequential patterns, respectively. Compared with this state-of-the-art method, our results achieve 99.98% accuracy, 90.46% F1 score, and 96.39% AUC, improving both model accuracy and generalisation capabilities while maintaining structural integrity.