Introduction of G-FDS
G-FDS is a fraud detection system reinforced with graph technology. Unlike standard FDS, G-FDS is specialized in tracing fraudulent activities within complex relationships and hidden patterns that are difficult to understand.
FDS is commonly used to detect and track down financial and tax-related crimes such as fake deposits, counterfeit cards, insurance frauds, subsidy frauds, and tax evasions.
This post will introduce the new Bitnine’s solution, G-FDS. The G-FDS uses AgensGraph as storage, which makes it different from any other fraud detection systems you have seen or heard before.
An AI-based G-FDS detects what existing FDS cannot. Let us look at the image below to see the difference between the G-FDS and general FDS.
Fraudulent transactions are identified based on a standardized pattern of data. A set of rules make it convenient to detect suspicious patterns and since the target is clear, the detection process can be done with a simple operation. However, the rule-based FDS cannot detect a pattern undefined by a set rule.
The advanced graph algorithm detects both the unstructured data and standardized patterns. Fraudulent activities are detected with the ‘motifs analysis’ technique. With this technique, it is possible to detect hidden meanings that deviate from the existing rules.
How G-FDS works
G-FDS analyzes the frequency of patterns and hidden relationships from numerous transactions accumulated from connected data. The process of applying the graph algorithm through motifs analysis is as follows.
Motifs are hypothetical patterns (M1, M2, M3…) that can be created with a certain number of nodes, also known as subgraphs. An example, ‘A transferred to B, and B transferred to C’ can be one of many patterns.
If a hypothetical pattern exists among normal and fraudulent activities, it is referred to as ‘motif fingerprint’. The number of repeated fingerprints within the total patterns are counted (Example: M4 = 24 times).
A score is derived from the number of motif fingerprints collected, which will be needed to find the pattern frequency of fraudulent activities. At this time, a score is given a statistical normal distribution, which is called a Z-score (average score). Based on the Z-score, patterns that frequently repeat in both fraudulent and non-fraudulent activities are listed and the differences are compared. If a motif fingerprint with a large difference in frequency is found among them, it is considered a suspicious pattern. (However, if there is a higher number of normal activities than fraudulent ones, it cannot be targeted as a suspect) The data of this suspicious pattern can be seen as fraud-type sample data and once this type of data is accumulated, it will be used to detect similar patterns and frauds in the future.
How far can G-FDS go?
There are three main areas in which G-FDS can detect crimes and frauds.
Detects and tracks down a fake deposit with a stolen name of a third party
Tracks the details and locations of transactions made with counterfeit or duplicated credit card
Detects fraudulent activities conspired by the perpetrator(s) and victim(s) to illegally claim insurance money
Tax Evasion / Money Laundering
Detects individual and corporate tax evasion crimes such as manipulation and concealment of incomes, transactions, documents, and properties
Subsidy & Tax Fraud:
Detects frauds that unlawfully receive public finances (tax returns, subsidies) through false reports
Detects activities of smugglers carrying contraband and drugs, etc. on behalf of the conspirators
Detects criminal activities in the e-commerce environment, such as customer account theft, product information leakage, illegal transaction, and hacking
Classifies and detects patterns of fake news that indiscriminately spread on social network services
Phishing / Smishing:
Detects fraudulent activities of stealing personal information via phone or text message pretending to be a financial institution
G-FDS Use Case: Exposing Fake Deposit
This case introduces tracking of a fake deposit under a stolen name of another person. This is an example of discovering a hidden relationship by linking a suspicious account with a number of clients’ various connections and authentication information with graph data. Based on the motif fingerprint, the fake deposit had a history of financial fraud in the past. By visualizing the graph pattern to facilitate real-time monitoring, it was possible to track another suspicious account related to the fake deposit.
Connect with Graph to see all crime signals!
G-FDS provides modeling services in the form of graph analysis. Since the ETL service facilitates the production of new data and transfer of existing data, the data used in the existing FDS can be applied with ease.
G-FDS can detect all areas that deviate from the existing rules and a combination of AI technology and graph DB enables preemptive response to crime and fraud. FDS with graph technology will become a new alternative at the point where the level of fraud in society as a whole, such as the financial sector, the insurance company, public administration, and manufacturing service sector, becomes more and more sophisticated.
If you have any questions, feel free to contact us.
Thank you! 😀