Online game abnormal transaction detection based on recurrent neural network using behavioral pattern

Abstract

Payment fraud in Online game is occurring frequently as real money trading market capable of selling game items in real money has emerged. In the previous research, a statistical based method to detect abnormal transactions was proposed. However, this method continues to misclassify some normal and abnormal users when their transaction statistics are similar with each other. We propose a sequence model that can overcome this problem. We also propose a new sequence model capable of reflecting purchase behavioral pattern between normal and abnormal users, so that it overcomes the problem of misclassification between normal and abnormal users with the same sequence information. In order to do that, we model the sequence of transactions with a recurrent neural network which also combines charge and purchase transaction features in single feature vector. In experiments using real data (a 483,410 transaction log) from a famous online game company in Korea, the proposed method shows a 78% recall rate with a 0.057% false positive rate. This recall rate is 7% better than current methodology utilizing transaction statistics as features.