Fraud Always Happens First Fraud patterns are constantly changing, and new methods of fraud are being introduced. Traditional anti-fraud methods cannot be upgraded in real time. There will be a time-consuming process of tag accumulation, calculation, analysis, and testing. This process often takes half a month and sometimes even several months. Fraud can be carried out in days or even hours. Anti-Fraud Detection Traditional anti-fraud measures have a long window of exposure. Now that fraud has become more industrialized, there are specialized intelligence centers that detect loopholes in anti-fraud systems. An anti-fraud model that has just been generated often quickly fails. There are usually two ways to increase the effectiveness of anti-fraud: Changing the structure of the model and Shortening the loss time. Unsupervised Machine Learning Redefines the Anti-Fraud Model Comparison of Unsupervised Anti-Fraud and Traditional Anti-Fraud Measures Among the various anti-fraud models shown above, blacklists and rules systems are still the most widely used methods at present. However, unsupervised machine learning is able to solve various problems in pattern recognition. The most important characteristic of UML is its ability to process and classify unknown data without the need for labels. The anti-fraud approach based on unsupervised learning is like a “virus and vaccine” relationship. Fraud can be compared to a virus with the ability to mutate. The application of an unsupervised anti-fraud system makes it possible to develop an antibody. Changes in the virus do not usually change the “antigen” so the antibody is still able to withstand the virus successfully. Unsupervised Machine Learning and Supervised Machine Learning Performance Comparison The above chart compares the performance of unsupervised machine learning to that of supervised machine learning. With simple maintenance, the anti-fraud effect achieved by an unsupervised machine learning algorithm can be maintained at a high level (blue curve). Since there is no need for the accumulation of labels and the setting of rules, unsupervised machine learning is a strong anti-fraud method. Unsupervised Machine Learning and Supervised Machine Learning Startup Contrast At the same time, due to the use of clustering without labels, unsupervised anti-fraud methods can achieve a fast start (blue curve), shortening the loss period. Unsupervised machine learning does not require manual marking, rule setting, or model training, and can be used to stop loss in a timely manner by detecting suspicious patterns exhibited in users’ behaviors. Author: DataVisor Technical Account Manager Huang Ying About DataVisor DataVisor is the leading fraud detection solution designed to uncover modern organized fraud attacks using DataVisor’s proprietary unsupervised machine learning (UML) approach. DataVisor excels at finding unknown fraud attacks by looking for suspicious correlation across all accounts without a need for prior fraud loss labels, and often before they can do damage. The DataVisor Platform provides advanced fraud management with accurate detection score, advanced fraud insights console and custom reporting, as well as flexible deployment and integration options. To date, DataVisor has processed over 800 billion events, detected 218M bad accounts, and protected over 4 billion user accounts from some of the largest financial institutions and online services in the world, such as Yelp, Pinterest, Alibaba and more. View posts by tags: Related Content: Stay up-to-date on the latest fraud insights and intelligence. Thank you for subscribing.