Boundary-aware dual-discriminator generative adversarial network for data augmentation in financial transaction fraud...

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Digital payments growth increases Financial Transaction Fraud Detection (FTFD) complexity.
• Extreme class imbalance, with few fraudulent cases, hinders FTFD model accuracy.
• Existing data augmentation methods struggle with anomalous fraud patterns and concealment strategies.

Purpose of the Study:

• To address the class imbalance issue in Financial Transaction Fraud Detection (FTFD).
• To improve the ability of FTFD models to accurately learn and detect fraud patterns.
• To enhance the quality of synthetic fraud data generation for better model training.

Main Methods:

• Proposed a Boundary-Aware Dual-discriminator Generative Adversarial Network (BADGAN).
• Integrated a boundary sample classifier and dual-constraint mechanism using distance adversarial learning.
• Enabled the generator to produce synthetic samples adhering to real fraud distribution and maintaining distance from decision boundaries.

Main Results:

• BADGAN effectively generates high-quality synthetic fraud samples near classification boundaries.
• The boundary-aware approach optimizes sample quality, improving downstream classifier performance.
• Experiments on real-world and public datasets confirmed BADGAN's superiority over peer methods.

Conclusions:

• BADGAN successfully mitigates class imbalance in FTFD.
• The proposed method enhances the accuracy and robustness of fraud detection models.
• BADGAN offers a promising solution for improving financial transaction fraud detection systems.