Product Overview
Problem
The Legislator obliged financial institutions to trace transactions and detect suspicious ones in order to report them to the Chief Inspector of Financial Information (CIFI).
Methods applied in systems serving to detect suspicious transactions are based on traditional SQL queries that operate within limits of earlier defined rules or parameters. At the same time they are not able to adjust to changeable patterns or schemes of “money laundering” without a considerable change to the program or interference of an expert in this field.
In case of present financial institutions which must process millions of transactions and hold thousands or millions of customers, systems based on pre-defined rules do not allow to be flexible in following still changing ways of “money laundering”.
Solution
FinDet system in detecting suspected transaction extends the assortment of existing methods with a technology of data mining and artificial intelligence.
The data mining process applied to detect and analyze huge quantities of data may be divided into a couple of steps. First, data must be integrated in order to form one format. The second step is visualization, conjunction, and interpretation of obvious and hidden connections, complicated relations, and patterns of activities included in data. The last step is getting results that shall be acceptable and understandable for a final user.
The Data Mining Module is based on the GhostMiner application and is composed of two parts: Developer and Analyzer.
Developer is used to automatically create classification rules and to detect suspicious transactions based on models prepared in the training set of given transactions.
Analyzer is used as a tool to detect suspicious transactions from among all transactions delivered by the system. Results of an analysis are demonstrated in a special summary report.
Application of the Data Mining Module allows detecting suspicious transactions in an automatic and more accurate manner. It requires to create own training set of transactions that defines classes of transactions. Such a set may be created in two ways:
Training transactions may be delivered from the register of suspicious transactions.
Training transactions may be also initially prepared due to the existence of the data clusterization module.
GhostMiner provides two ways from among mostly known data mining methods serving commonly to detect suspicious transactions:
- clusterization
- classification
Algorithms applied in the GhostMiner package are based on the most modern methods of neural networks, decision trees, and visualization methods.
