Design And Implementation Of High Risk Fraud Identification System

The research objects of this paper are accounts or equipment of users in PC and mobile terminals. By collecting data of users'accounts like operation behavior, transaction process, equipment operation trajectory and trading state of equipment, and utilizing static rule and data analysis training, the system finds out whether it is a fraud. If so, the user would be classified as a risky consumer. The front-end system would show the fraud index of the risky user, which is advantageous for relevant personnel to make a decision.The high risk of fraud system introduced in this paper is based on actual business needs of E-commerce, game platforms, small-and-medium-sized center for responsible lending, bank and so on. It could evaluate suspicious degree of enterprise users, find suspicious accounts from a large number of users and devices, and mark them with corresponding degrees. It provides both automatic and artificial ways to recognize suspicious objects. At the same time, it offers unified management to systems like scoring system, feature extraction engine, and messaging access system and so on. It also conducts monitoring management and collaborative scheduling to system services by integrating Zabbix[1]and Zookeeper. According to actual business acquirement, this paper uses UML to do requirement analysis, and it introduces system requirements in detail by utilizing use case diagram in UML. During the overall framework design, the paper uses MVC framework and Python+Django to design overall framework and Web system respectively according to system acquirement, in the meanwhile, it realizes the back-end logic of the anti-fraud systems by utilizing Scala+Akka+Flume+Kafka+MongoDB, and the display of Web front-end is realized by React+Flux+Bootstrap.This paper elaborated the core functioning modules and overall framework design in the part of detailed design and implementation. It uses Akka[2] parallel framework technology to realize Compositor, Feature Engine, Scorer and describes the detail of data processing and co-scheduling among modules. In the level of system deployment, this paper prevents system from being unable to work when one server crashes by using Zookeeper to co-schedule subsystem and conducting double online backup[17]. The solution to duplex server crash is also included[3].Through a lot of simulation data and real data test, the system itself has reached a relatively stable state. The overall performance of the system, the load has support around 200/sec processing speed and functions between the various systems can be well coordinated streaming ? Under the different fraud scenarios and features of different extraction system model, with the scoring system and obtain external library scoring model and game on behalf of the charge and the promotion of arbitrage scene fraud recognition rate has reached 95% (Note:the data comes from a large number of test results).