Importance of behavioral anomalies in detecting digital fraudsters at the verification stage

Digital fraud is becoming increasingly complex and sophisticated, and is also growing in volume, which requires improving the mechanisms for protecting digital services. The article discusses the importance of behavioral anomalies for identifying fraudsters at the verification stage. It presents the results of an experiment using the gradient boosting model (Catboost). While analyzing the data of 100,000 users, we have identified 2863 of them as fraudsters. The key behavioral factors, anomalies in which indicate possible fraud, are document loading time, time to complete the biometrics stage, time to fill out the questionnaire, the number of attempts to load documents and biometrics, the average pause between filling in the form fields, and the reaction time when starting the biometrics stage. The results show that the behavior of fraudsters often deviates from the average for these indicators. These data can be effectively used for adapting verification systems in order to minimize losses among honest users, maintaining at that a high level of protection. The limitations of the study are associated with the analysis of only the verification stage, which emphasizes the need for further study of user behavior after this step. The findings emphasize the need for a comprehensive approach to monitoring user behavior to combat digital fraud.