Making Sense of Data: A Practical Guide to Exploratory Data Analysis and Data Mining

7.6 OTHER METHODS

There are many methods for building both classification and regression models. The following section briefly describes a number of alternative approaches. More details on these approaches are referenced in the further reading section of this chapter.

• Multiple linear regressions: The method described for simple linear regression can be extended to handle multiple descriptor variables. A similar least squares method is used to generate the equation. The form of the equation is y = a + b₁x₁ + b₂x₂ + . . . + b_nx_k where y is the response, x₁ to x_n are the descriptor variables, a is a constant, and b₁ to b_n are also constants. For example, when attempting to predict a potential customer's credit score (CS) a multiple linear regression equation could be generated. The equation could be based on the number of missed payments to other credit cards (MP), the number of years with no missed payments (NMP), and the number of good standing loans (GSL), as for example in the following equation:

  Table 7.19. Optimization of neural network

  

  

  

  

  CS = 15 – 18 × MP + 12 × NMP + 10 × GSL