Abstract

Fraud in the airline industry leads to substantial financial losses, amounting to billions of dollars annually. While fraud detection systems are more commonly deployed in the banking sector, the aviation industry faces increasingly complex and damaging fraud cases. With the exponential rise in fraud incidents, airlines are experiencing significant financial strain. Traditionally, rule-based systems have been the primary defense against fraud in aviation. However, their limitations, especially in combating novel or zero-day fraud attacks, have driven the need for more sophisticated solutions using machine learning. Common types of fraud in this sector include stolen credit card transactions, chargeback abuse, loyalty program manipulation, and ticket scams.

To address these challenges, we developed a unified fraud detection system leveraging Transfer Learning and Federated Learning, which has demonstrated exceptional performance in detecting fraud. Initially, the model showed strong results, achieving 95% accuracy for stolen credit card transactions, 98% for chargeback abuse, 93% for loyalty program fraud, and 96% for ticket scams. After implementing federated learning, the system’s performance further improved. The accuracy for stolen credit card transactions increased to 96%, chargeback abuse reached 99%, loyalty program fraud improved to 94%, and ticket scams rose to 97%. These results demonstrate the effectiveness of combining transfer learning and federated learning in creating a scalable, accurate, and privacy-preserving fraud detection solution tailored for the airline industry.

Keywords: Airline fraud detection; Transfer learning; Federated learning; TabNet; BERT; Stolen credit card fraud; Chargeback abuse; Loyalty program fraud; Ticket scams; AI fraud detection; Privacypreserving ai; Anomaly detection in airlines; Machine learning in airlines; Natural language processing in fraud detection