Project Description
Automated fraud behaviors detection on electronic payment platforms is a tough problem. Fraud users often exploit the vulnerability of payment platforms and the carelessness of users to defraud money, steal passwords, do money laundering, etc, which causes enormous losses to digital payment platforms and users. There are many challenges for fraud detection in practice. Traditional fraud detection methods require a large-scale manually labeled dataset, which is hard to obtain in reality. Manually labeled data cost tremendous human efforts. Besides, the continuous and rapid evolution of fraud users makes it hard to find new fraud patterns based on existing detection rules. In our work, we propose a real-world data oriented detection paradigm which can detect fraud users and upgrade its detection ability automatically.
Model Overview
This is the overview of our detection model. Our model should contain two main functions: building fraud users blacklist and automatically updating the blacklist. The whole detection model contains three phases:
Phase I: Merging features from raw data. Platforms collect two main classes of information: operation data and transaction data. We first merge the two classes data, and the merged features are passed to our fraud detection part.
Phase II: Building fraud users blacklist. Merged features in Phase I are of high dimensions, which can not be used directly. We can omit ineffective and noisy features and get efficient low dimension features with the help of adversarial autoencoder. In the meantime, we can detect fraud users by a few labeled data. It is a semi-supervised learning process. Detected fraud users will be added in the blacklist.
Phase III: Updating fraud users blacklist by cluster analysis. The key idea is to find new fraud users beyond existing detection rules. In this phase, we train a new AAE network without labels to learn the latent representations of users. Then we cluster these latent variables from the network. After clustering latent variables, different users groups are formed. We detect fraud users with new patterns from these fraud groups.
The details of our research is shown in our paper FraudJudger: Real-World Data Oriented Fraud Detection on Digital Payment Platforms (arxiv) and FraudJudger: Fraud Detection on Digital Payment Platforms with Fewer Labels (ICICS 2019).
This is a project cooperating with tiancheng.
