ABSTRACT
At this point you have had an opportunity to gain some experience using the data mining tools contained in R and R Commander. Given this basic background, the primary goal of this chapter is to provide you with a framework that will allow you to rapidly develop a reasonably good model. A model can usually be improved by devoting more effort to it, but there are diminishing marginal returns to this effort. The popular 80/20 rule applies, and the “reasonably good” model is sometimes referred to as the “80% model,” implying that with 20% of your modeling time and effort you will get the first 80% of the return, and it will take 80% of your time to get the remaining 20% return (that results in the best possible model). Before presenting our rapid model development framework, we first present an additional tool, stepwise variable selection (Venables and Ripley, 2002). Stepwise variable selection is an automated way of finding the set of predictor variables in a regression model (either linear or logistic) that is the best based on some criterion (often AIC). A tutorial showing you how we applied the framework to a new database follows the presentation of the rapid development framework. At the end of the tutorial we will show how to use R Commander to “score” a database. Scoring a database amounts to creating a new variable in a database that allows the user to rank each customer in the database in terms of their attractiveness as a campaign target based on the fitted values of a particular predictive model. We are already familiar with this idea: scoring is exactly what is required to create a lift chart, but the score is not added to the database.
