Detecting Online Game Chargeback Fraud Based on Transaction Sequence Modeling Using Recurrent Neural Network

Abstract

We propose an online game money chargeback fraud detection method using operation sequence, gradient of charge/purchase amount, time and country as features of a transaction. 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.