Data Mining Algorithms: Explained Using R

Chapter 5

Linear classification

5.1 Introduction

The linear model representation is a special case of the parametric representation which assumes that the model's predictions are calculated by applying a representation function to attribute values and a set of real-valued parameters. This is particularly natural and extremely common for regression models which make real-valued predictions. The same approach can also be adopted to represent classification models, though. Moreover, such models can be created by the same or nearly the same algorithms as those that normally deliver regression models. This can be achieved in several ways, some of which are discussed in this chapter. The chapter will focus on issues related to adopting parametric regression methods to the classification task. This is essentially based on using a composite model representation function, consisting of a real-valued inner representation function and a discrete outer representation function that assigns class labels based on the former.

According to this book's task-oriented organization, chapters devoted to classification algorithms precede those covering regression algorithms. A book must have a linear structure and of different possible presentation orders; this one is believed to provide the best didactic value. For consistency, this linear classification chapter appears before Chapter 8 devoted to linear regression. However, the linear model representation and the corresponding modeling algorithms ...