DECISION TECHNOLOGIES IN DATABASE MARKETING: PART VII
by Gene M. Ferruzza, Senior VP, Decision Technologies
Non-Parametric Models
Non-parametric modeling techniques also are data-driven. In contrast to parametric models, non-parametric models require no assumptions about the functional form of the relationship between the target behavior (dependent variable) and the independent variables. In cases where it is difficult to fit the data to particular form, non-parametric techniques offer more flexibility. In other words, they have the ability conform their structure to that of the best solution. Non-parametric technologies vary in their flexibility, and particular techniques are best suited to particular types of marketing applications. Some of the more popular non-parametric technologies are neural networks, advanced statistical clustering techniques, and inductive decision tree technologies. Particularly popular are backpropagation neural networks, K clustering techniques, fuzzy logic, and rule-based decision tree technologies, such as ID3, C4.5, CHAID, and case-based reasoning (CBR) technologies.
Rule-Based Segmentation
In data mining, a "rule" is a method of extracting or segmenting, transforming, summarizing, or even reporting information from a database. Rule-based methods are the most widely used approaches in data mining because they are relatively easy to develop, are easily deployed, and are easily understood by most marketing users.
Rule-based segmentation methods are the most versatile data-mining techniques in database marketing.
Rules can be generated by fixed, parametric, or non-parametric models and may be developed by domain experts or derived through data-driven processes. Regression models often are converted to "score cards," which are nothing more than a set of rules extracted from the model. Some of the non-parametric techniques mentioned above generate sets of rules easily used in database marketing.
Rules are made up of simple inferences in the form of "if-then" statements: if this is true then do something. In this statement, [this is true] is considered a premise, and [do something] is considered the result.
When a premise is created, operands and operators are used to determine whether the statement is true or false. If the statement is true, then the result will be executed; if the statement is false, the result will not be executed. These are common operands:
These are some common operators:
Rule-based segmentation methods can be categorized in two ways. First, they can be categorized by the type of data being used (i.e., customer behavior data, primary research data, or both). Second, they can be categorized by whether the segmentation is static or dynamic. "Static" segmentation refers to creation of fixed, unchanging customer segments. Customers can move from one segment to another, but the segments always remain the same. "Dynamic" segmentation refers to creation of customer segments that may change at any time, depending on the goals for the marketing program. In addition, dynamic segmentation approaches allow the simultaneous use of many segmentation schemes and strategies, which may compete with each other in accuracy.
The simplest form of a rule is a query on a database. Rules usually are
generated by expert users of the data or by business domain experts. If a
bank marketer, through experience, believes that the bank's most profitable
customers are over the age of 50 and have annual incomes of over $75,000, he
can query his data mart for the most profitable customers, as defined by age
and income. So is this data mining? What knowledge has been discovered?
---
Gene Ferruzza may be contacted at gmf@cmsnet.com
---
The eighth installment in this series will appear in next week's D S * .