The operator «VimpelCom» (Beeline brand) has developed a neural network that helps to identify fraudulent schemes two to three hours before the transfer of money. The system has been successfully tested and is already being used in major Russian banks.
The Beeline team has updated its approach to fraud analysis by creating a new algorithm that recognizes signs of prolonged exposure of fraudsters to clients. Experts identified eight key features that indicate possible fraud. The algorithm analyzes user activity and compares it with familiar models, identifying deviations, such as an increase in contacts with new people.
When a bank receives a request for a suspicious transaction, it can refer to Beeline's data. If the system detects signs of fraud, it signals the security service of the bank to suspend the transaction to warn the client.
The updated system has already proven its effectiveness: in tests on 4 million clients, the accuracy of detecting fraudulent activities was about 92%. The algorithm was able to warn about 18 thousand cases of fraud.
Read also on the topic:
Yandex neural networks began to find fraudsters seven times faster