Researchers at Massachusetts Institute of Technology (MIT) in the US have devised an artificial intelligence (AI) method to reduce false positives during detection of credit card fraud.
It is expected that the new approach will minimise declining of legitimate transactions, in turn saving money for banks and avoid inconvenience for customers.
Machine learning has been in use for financial fraud detection since early 1990s. Such models are trained to extract behavioural patterns from past transactions (features) that indicate fraud.
Upon swiping, the card pings the model and in case the features match fraud behaviour, the sale will be blocked.
The researchers noted that one reason for more false positives is that fraud-detecting technologies adopted by a bank incorrectly flag the sale.
To address such issues, the team devised an ‘automated feature engineering’ method called Deep Feature Synthesis (DFS), which extracts more than 200 detailed features for each transaction.
The technique is based on ‘primitives’, simple functions that take two inputs and give an output. These primitives are responsible for analysis and determination of a transaction as fraud or non-fraud.
MIT principal research scientist at Laboratory for Information and Decision Systems Kalyan Veeramachaneni said: “The big challenge in this industry is false positives. We can say there’s a direct connection between feature engineering and false positives. That’s the most impactful thing to improve accuracy of these machine-learning models.”
When tested on a 1.8 million transaction dataset, the new model was able to decrease false positive predictions by 54% than conventional models. It is estimated that this cut down saved the bank around $220,000 in lost revenue.