A Deep-forest based approach for detecting fraudulent online transaction

Abstract

Abstract Fraudulent transaction is the one of the most serious threats to online security nowadays. Artificial Intelligence is vital for financial risk control in cloud environment. Many studies had attempted to explore methods for online transaction fraud detection; however, the existing methods are not sufficient to conduction detection with high precision. In this chapter, we propose a Deep-forest based approach for online transactions fraud detection, which integrates differentiation feature generation method and deep-forest based model. As a single-time transaction's information, which does not contain information such as the user's behavior, is not sufficient for detecting fraudulent transaction, we introduce a transaction time-based differentiation feature generation method into our scheme. Individual Credibility Degree (ICD) and Group Anomaly Degree (GAD), which are based on transaction time, are derived to distinguish between legal and fraudulent transaction. Furthermore, to deal with the extreme imbalance of online transactions, we apply Deep-forest algorithm to detect fraudulent transactions. While raw deep-forest model could ignore the outlier transaction samples, we enhance the raw Deep forest with detection mechanism for outliers, paying more attention on outliers to promote the precision of fraud detection model. Finally, we conduct test using one bank's transaction data. Compared with random Forest-detection model, our method improves precision rate by 15% and recall rate by 20%.