As global trade connectivity increases, so does import fraud. Facing vast trade volumes, customs can only inspect a fraction of declarations. We developed AI algorithm GraphFC, which requires minimal labeled data for training and accurately detects illegal import declarations. In collaboration with WCO, GraphFC has been tested in several African countries. The results were presented at AI top conference ACM CIKM 2023, showcasing Taiwan＇s significant contribution to AI in customs fraud detection.

(1) Utilizing advanced graph neural networks, GraphFC analyzes customs import data with minimal labeled input to accurately predict illegal transactions, allowing focused inspections on high-risk cases.
 
 (2) Through self-supervised learning and information diffusion, GraphFC adapts to label scarcity, enhancing its ability to generalize and quickly adapt to new market entrants.
 
 (3) GraphFC enables it to continuously learn from new data, adjusting to the evolving international trade environment.

GraphFC is currently deployed in Nigeria and Malawi＇s customs systems and is being promoted to all WCO member countries. It aims to impact globally by: Detecting customs fraud to ensure lawful trading, Increasing tax revenue and minimizing tariff evasion losses, and Extending to the banking sector to spot illegal financial transactions and prevent money laundering, leveraging its open-source nature. GraphFC protects customs inspectors by reducing exposure to risky environments.