Machine Learning for Loyalty Program Fraud Detection
Challenge
One of the largest airlines in Europe was seeing a significant increase in fraud in its frequent flyer loyalty program.
As scammers used more sophisticated technologies, the fraud detection team was not able to shut them down early.
The airline’s Chief Data Scientist needed a way to increase his team’s detection capabilities without increasing the size of the team.
Solution
Data scientists built a complex fraud detection model looking for patterns across a wide variety of variables.
Machine learning (ML) tools were used to train the model. They assessed the model’s fairness until they were confident that the results were meaningful and fair.
Once operationalized, the model automatically retrains when code or data changes, making it smarter over time.
Results
The ML model identifies fraud with 99% accuracy.
Data scientists eventually simplified the ML model to reduce preprocessing costs by 20% to increase productivity without losing accuracy.
The fraud detection team is better equipped to identify small-scale fraud which is more difficult to detect. They have increased the detection of small fraud transactions by 300%.s