Fujitsu Laboratories and LARUS Business Automation S.r.l. (LARUS)(1) have jointly verified that credit card payment fraud can be detected with high accuracy by integrating Deep Tensor(2), an explainable graph AI technology(3) developed by Fujitsu Laboratories into the LARUS platform for graph databases.
Fujitsu and LARUS achieved this by linking the LARUS platform for graph databases with Fujitsu’s graph AI technology. Compared with previous, rule-based approaches created manually by data analysts, the fraud detection rate improved from 72% to 89%, while the false detection rate was successfully reduced by 63%. Additionally, it was confirmed that the creation of rules for fraud detection could be supported by presenting the decision factors of fraud cases detected with graph AI technology. Going forward, both companies will verify the effectiveness of this technology in other industries with the objective of delivering practical uses for graph databases and graph AI.
Detailed results of this verification trial will be demonstrated at the “AI & Big Data Expo Europe 2020” conference, which will be held online from November 23rd, 2020 (Monday) to 24th (Tuesday) Central European Time.
In recent years, there have been growing expectations around the use of graph data in AI applications in various fields and industry verticals, including the analysis of SNS activity history, representation of chemical molecules, financial transactions and tracking of virus infections. Instead of utilizing conventional relational databases(4), storing this kind of real-world relationship into a graph database, which underscores the relationships between underlying data elements, enables expression of data elements relationships directly, thereby enabling advanced analysis and discovery of new insights into real-world scenarios.
In the field of finance, for instance, it’s possible to extract important information utilizing graph databases by analyzing the relationships between transactions. Analysis of individual transactions alone remains insufficient, especially for detecting complex, fraudulent transactions, and all transactions must be analyzed together as a graph structure. For example, in the case of self-financing, a type of fraudulent transaction, even if individual transactions appear normal, when the relationship of the different transactions are analyzed together, circular or loop like patterns may become apparent. Graph databases are well suited for detecting fraud from graph patterns like these loops.
To date, there has been a limit to how much data analysts have been able to create rules for fraudulent trading patterns, and concerns persist around the risk of misidentification and false detection. Consequently, it has become necessary to further streamline graph AI technology.
Overview of the Verification Trial
In this trial, table data containing details of individual transactions was converted into graph data expressing the relationship between different data elements. These were subsequently analyzed by combining the platform for graph databases which LARUS provides with Fujitsu Laboratories’ “Deep Tensor” technology. Fujitsu and LARUS used actual credit card data and POS data and verified the degree of improvement with fraud detection rate or false detection rate by comparing with manually created fraud detection rules. In addition, by utilizing the “explanation of detection with visualization” functionality, which is another important feature of “Deep Tensor,” it was possible to show the reasoning behind the different decisions to the satisfaction of data analysts.
In the verification of credit card transaction data from a payment services provider, fraud detection rates improved from 72% to 89% and the false detection rate was reduced by an average of 63%, when the new graph AI based approach was used in comparison with the conventional manual, rule-based approach.
Furthermore, Fujitsu also confirmed that the explanation of detection with visualization was adequate from the viewpoint of the data analyst, making it possible to support the creation of new rules for improved fraud detection. (Figure 2).
In the future, both will additionally verify the technology’s effectiveness with data drawn from different industries, paving the way for the practical use of graph data and graph AI in a broader range of application areas.
Fujitsu, the Fujitsu logo and “shaping tomorrow with you” are trademarks or registered trademarks of Fujitsu Limited in the United States and other countries. Other company or product names mentioned herein are trademarks or registered trademarks of their respective owners. Information provided in this press release is accurate at time of publication and is subject to change without advance notice.
Date: 24 November, 2020
City: Kawasaki, Japan, and Venice, Italy
Company: Fujitsu Laboratories Ltd. ,LARUS Business Automation S.r.l.