Regression Analysis - Visual Data Mining: The VisMiner Approach - page 147

Figure 6.7 ANN Regression

curves on the opposite edges of the surfaces. For example, look at the curve along
the X (bathrooms) axis where sqFeet ¼ 500. It shows a negative contribution of
each additional bathroom. In other words, for a small home, squeezing in
more bathrooms reduces its value. Look at the curve along the opposite edge
(sqFeet ¼ 5,000). It shows a positive contribution for the first four or five bath-
rooms, then a negative return for the sixth. When the shapes of the curves at
opposite edges of the surface differ, the modeler has detected interaction between
the input attributes. We will see more dramatic differences in later models.
Given the relatively low levels of R 2 achieved by the modelers using sqFeet
and bathrooms, the question should be asked as to which additional input
attributes would improve the fitness of the model or if some other changes could
be made.
Return to the correlation matrix to identify the top four attributes with respect
to correlation with price.

Create a derived dataset named “best4” consisting of price and the four
most correlated attributes: sqFeet (0.713), bathrooms (0.476), lot (0.414),
and bedrooms (0.365).