One of the largest bank having multiple branches in the North America.
To detect and prevent credit card fraud in real time, the client wanted to have an intelligent module to be incorporated into their system. Also, they wanted to have the system to be intelligent enough to upgrade itself with new fraud patterns ensuring that it doesn’t become outdated with change in credit card fraud trends.
VOLANSYS created a machine learning model capable of detecting credit card fraud in real time. Also, deployed it using Google TensorFlow Serving, so that it can be trained in real-time and model can be updated smoothly in run-time without impacting the credit card fraud detection mechanism.
- About 20 TB of data for last 5 years
- Customers transaction Information like transaction id, amount, paid to, date and time, transaction type etc.
- Transaction data like the type of transaction, amount, user location, time etc.
- User complaints data containing, information like type of fraud, fraud amount, date & time, fraudulent activity location etc.
Visualized data and found the normal and abnormal transaction pattern which indicated more than around 75% of credit card fraud and fraud attempts were made from mainly 3 countries
- Identify the key features that play important role in identifying normal and abnormal transactions
- Identify the transactions, that appear to be abnormal, but were actually normal transactions
With an accuracy of 98%, predicted the probability of a credit fraud transaction based on fraud analysis and behavioral analysis of user activity data for the transaction
- Based on available data predict the probability percentage of types of credit card fraud transaction in future
- Calculated and provided percentage loss based on available data that may be incurred by the Bank in case of credit card fraud in future
- Maximum number of credit card fraud attempts were from mainly 3 countries
- Maximum credit card fraud was done in airlines transactions i.e. 50% and the second highest were done in money transfer i.e. 17%
- If the model predicts transaction to be a fraud with probability of more than 70%, implement user registered mobile based OTP authentication to complete transaction
- Block card and send customer notification if a model predicts transaction to be a fraud with probability of more than 70% and user OTP authentication fails
- Real-time automated fraud detection mechanism with reduced prevention action time
- More secure mechanism to prevent credit card fraud
- Increase in customers reliability of bank security