Investigative Data Mining for Security and Criminal Detection

8.11 Building a Fraud-Detection System

To demonstrate how a fraud detection system can be constructed using advanced data mining technologies and tools, this sample will be presented with the use of a data set from an actual e-business that sells electronic consumer products via its Web site. For this case study, we will go through several processes, including the following:

• Assembling samples of transactional data

• Enhancing the customer information with offline demographics

• Visualizing associations of fraudulent transactions with a link analyzer

• Mapping features of fraudulent transactions using a SOM

• Constructing predictive models for identifying fraud cases via neural networks

• Creating decision trees and extracting conditional rules via machine-learning algorithms

• Building an ensemble of models for detecting fraudulent transactions in real time