Method and system for detection of fraudulent transactions

摘要

Method/system for automated detection of fraudulent/suspicious credit card transactions. The system collects historical data of customers' transactions over a specified time period. The system identifies customers similar in spending behavior over the specified time period and clusters them into groups. The system detects suspicious transactions by behavior deviations from a normal behavior of a given cluster of customers. The system implements two algorithms for combining the customers into groups (clusters). The first algorithm is based on customers' relationships with the same sellers. The second algorithm is based on similarity of customer spending-based categories of sellers. The execution of clustering algorithms results in creation of two types of experts. The initial data used in the system comprises a teaching set of transactions (filtered and modified historical data over a time period), fraudulent transactions discovered over a given time period and ongoing sets of transactions for detection of fraudulent transactions.