8.16 Clustering Fraud


8.16 Clustering Fraud

We next move to another type of visualization analysis that at the same time performs an autonomous clustering of the data using a SOM neural network from SOMine, a data mining firm. As we found in Chapter 6, a SOM is a neural-network architecture that allows for unsupervised learning and is used to perform clustering of data, allowing it to organize itself around similar segments. For this situation, we wanted to discover the similarities of fraudulent transactions along different information factors. A SOM analysis is used most commonly in exploratory situations, where little is known about the data set. Using a binary value classification system, where a 0 (light) represents a paid transaction and a 1 (dark) is a fraudulent transaction, we are able to generate the two-dimensional clustering map shown in Figure 8.2.

click to expand
Figure 8.2: A clustering map where light shades are legal and dark areas are fraudulent transactions.

Using this SOM analysis, sections of the clustering map can be marked and extracted from the database into a subset for further analysis. In Figure 8.3 we marked the dark sections that represent fraudulent transactions at the right mid-section of the large map.

click to expand
Figure 8.3: We mark the section of fraudulent transactions.

The section marked from the clustering map can now be cut and pasted into a spreadsheet for a more detailed review. For example, we may want to view the fraud transactions according to products, average price ranges, or demographics (see Figure 8.4).

click to expand
Figure 8.4: Camcorders with an average price of $1,052 are a major target for fraud.

Through this clustering analysis, specific sectors of transactions related to fraud can be identified and extracted in order to develop a profile of shoppers likely to be responsible for criminal activity. For example, additional exploratory analyses can be performed by these types of visualization tools to discover associations between fraudulent transactions and other customer clickstream behaviors and demographic features. These types of analyses can yield actionable insights, enabling merchants to become more knowledgeable about the characteristics of criminals on their Web sites.




Investigative Data Mining for Security and Criminal Detection
Investigative Data Mining for Security and Criminal Detection
ISBN: 0750676132
EAN: 2147483647
Year: 2005
Pages: 232
Authors: Jesus Mena

flylib.com © 2008-2017.
If you may any questions please contact us: flylib@qtcs.net