View Full Version : real time pattern matching using data mining

07-22-2008, 10:09 PM
i am new to the concept of data mining and pattern matching and have read several articles re wega and its alogoritms etc.

i am lookig at how we can enable a credit card fraud detection system from ecommerce sites that while it may go to a rules engine to check some variables , how we can use the data mining patterns for customers and feed back some of the results into the client profile database so next time they can be checked more rigorously.

i have read that we need to create .aarp files with data and defintion and run this through.

has anyone got at some small example of how pentaho could be used for this.

many thanks:)

07-23-2008, 12:24 AM

Pentaho data mining (Weka) can be used for this. The basic process is to use historical records of fraudulent and non-fraudulent credit card usage to train a predictive model. The model can then be used to assign a score (probability) of a case (pattern of usage) being fraudulent. The key to this is to define features that are likely to be predictive of fraudulent use. Pentaho has tools that can be used to convert data stored in databases into the flat ARFF file format that Weka uses.

Once a model is trained and validated, it can be deployed with Pentaho data integration (PDI) to update a client profile database with risk scores.