AI4Risk/antifraud
A repository for financial fraud detection
What it solves
AntiFraud is a unified research framework designed to detect financial fraud. It provides a standardized way to implement, train, and evaluate various deep learning and graph-based fraud detection methods, reducing the need to build separate pipelines for different algorithms.
How it works
The framework integrates several representative fraud detection approaches, including convolutional neural networks (CNNs), spatio-temporal attention networks, and Graph Neural Networks (GNNs). It uses a unified data processing pipeline to handle datasets like YelpChi, Amazon, and S-FFSD, including feature engineering and the construction of high-order transaction graphs.
Who it’s for
Researchers and data scientists specializing in financial security and fraud detection who need a benchmark framework to test and compare different deep learning and graph-based models.
Highlights
- Multiple Model Implementations: Includes a wide range of methods such as MCNN, STAN, STAGN, GTAN, RGTAN, HOGRL, and Grad.
- Unified Pipeline: Standardized data processing, training, and evaluation pipelines for consistency across different models.
- Graph-Based Approaches: Strong focus on graph representation learning and graph augmentation for fraud detection.
- Comprehensive Benchmarking: Provides results across multiple datasets to allow for direct performance comparison.
Related
- Project
- Project
- Project
- Project