New Transformer Architecture Enhances Financial Fraud Detection

Researchers have developed the Multi-Stream Fraud Transformer (MSFT), a novel architecture designed to detect financial fraud by analyzing heterogeneous event streams like transactions and login sessions. The MSFT utilizes independent Transformer encoders for each stream and employs configurable fusion mechanisms to combine their representations. Experiments on a large dataset showed that MSFT significantly outperforms traditional gradient-boosted trees and single-stream Transformer models, achieving an AUROC of 0.9961 with time-aware positional encoding. The study also indicated that gated fusion is optimal for production deployment due to its high precision, and that risk event streams are the most indicative of fraud. AI

IMPACT This research introduces a more effective method for detecting financial fraud, potentially improving security and reducing losses in digital banking.

RANK_REASON The cluster contains a research paper detailing a novel model architecture and its performance on a specific task. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• Mohammadamin Dashti Moghaddam
• Multi-Stream Fraud Transformer
• Transformer
• XGBoost

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →