Abnormal Detection of Cash-Out Groups in IoT Based Payment

Area of Science:

• * Financial Technology (FinTech)
• * Cybersecurity
• * Data Science

Background:

• * The rise of online/mobile transactions, particularly in the Internet of Things (IoT), has increased opportunities for sophisticated fraud schemes like cash-out groups.
• * Traditional detection methods struggle with the hidden nature and larger transaction volumes of cash-out groups, especially in regions with favorable fee structures.
• * Identifying and mitigating unlawful credit card cash-out operations is crucial for financial security.

Purpose of the Study:

• * To develop and validate a novel method for detecting cash-out groups in IoT-based online payment systems.
• * To improve the efficiency and accuracy of identifying fraudulent financial activities.
• * To provide a framework for classifying the severity of detected groups for risk management.

Main Methods:

• * Seed card identification and diffusion using Approximate PageRank (APR) for local graph clustering.
• * Construction of a merchant association network using Node2Vec graph embedding on suspicious cards.
• * Clustering of merchants in Euclidean space via DBSCAN and subsequent severity classification.

Main Results:

• * The proposed method successfully identified 74.4% (145 out of 195) of known cash-out groups across four banks.
• * An additional 178 cash-out merchants were identified within the same acquirers, totaling 30,586 merchants.
• * The framework demonstrated significant effectiveness in uncovering hidden fraudulent networks.

Conclusions:

• * The developed method provides a robust solution for detecting sophisticated cash-out groups in online payment environments.
• * The approach enhances the capabilities of financial institutions in combating large-scale fraud.
• * The framework has been adopted into a real-world cash-out group detection system by a major Chinese payment processor.