美联储-CardSim_支付卡欺诈检测研究的贝叶斯模拟器（英）-2025_41页_1mb

CardSim: A Bayesian Simulator for Payment Card Fraud Detection Research

Summary by Report Analysis Expert

1. Background

• Payment card fraud is rising globally, exacerbated by generative AI tools enabling sophisticated fraud tactics.
• Publicly available payment transaction data is scarce due to privacy and economic concerns, limiting research reproducibility.

2. Key Contributions

• CardSim: A flexible, scalable Bayesian simulator for generating synthetic payment transaction data and fraud patterns.
• Methodological Features:
  • Bayesian Approach: Uses Bayes’ theorem to associate payment features with fraud probability.
  • Modular Design: Easily adaptable to changing payment and fraud trends using publicly available data.
  • Data Privacy: Avoids reliance on sensitive raw data by leveraging statistical distributions.

3. Methodology

1. Simulator Workflow:

   • Payer/Payee Characteristics: Derived from surveys (e.g., Diary of Consumer Payment Choice).
   • Transaction Simulation: Generates payment attributes (amount, type, location, distance, time) based on calibrated distributions.
   • Fraud Labeling: Uses Bayesian ranking to assign fraud flags, targeting extreme class imbalance.

2. Parametric Flexibility:

   • All parameters adjustable, allowing users to simulate evolving fraud behaviors.
   • Computationally efficient (e.g., 3.16 million records >10 seconds).

4. Applications

• Machine Learning Testing: Evaluates fraud detection models (e.g., logistic regression, XGBoost, neural networks) on synthetic datasets.
• Interpretability: Uses SHAP framework to explain model predictions and identify key fraud indicators.
• Performance Metrics: Demonstrates effectiveness in precision, recall, and F1 score under imbalanced data.

5. Limitations and Future Directions

• Current Constraints:
  • Simplifies fraud types (no typologies like spending sprees).
  • Lacks real-world authorizations and behavioral analytics (e.g., recency/frequency).
• Future Enhancements:
  • Incorporate fraud typologies and unsupervised learning.
  • Extend to other payment systems (e.g., ACH, wire transfers).

6. Conclusion

CardSim addresses data scarcity in payment fraud research by providing a robust, modular, and privacy-preserving simulation framework. It enables rigorous testing and evaluation of fraud detection models, advancing research and practice while highlighting areas for future development.