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