2. Mapping Regression Into Classification

In regression problems we are given samples of a set of independent (predictor) variables x1, x2, …, xn , and the value of the respective dependent (output) variable y³. Our goal is to obtain a model that somehow captures the mapping y = ƒ( x1, x2, …, xn) based on the given samples. Classification differs from this setup in that the class is categorical instead of numerical.

Mapping regression into classification is a kind of pre-processing technique that enables us to use classification algorithms on regression problems. The use of these algorithms involves two main steps. First there is the creation of a data set with discrete classes. This step involves looking at the original continuous class values and dividing them into a series of intervals. Each of these intervals will be a discrete class. Every example whose output variable value lies within an interval will be assigned the respective discrete class. The second step consists on reversing the discretization process after the learning phase takes place. This will enable us to make numeric predictions from our learned regression model. Figure 1 shows a diagram of this process:

• 2.1 Methods For Splitting A Set Of Continuous Values
• 2.2 Making predictions from the learned models