Databases Reference
In-Depth Information
Figure 6.8
ANN Regression Interaction
same as that third bathroom in a six-bedroom home? In other words, does the
value of a bathroom change depending on the number of bedrooms in a home?
To detect interaction:
View a regression model of the ANN model.
To facilitate comparison, view a regression model of the third-order
polynomial model to the side of the ANN model.
On both plots, select sqFeet for the X axis and lot for the Z axis.
We previously presented the concept that interaction exists between input
attributes when the shape of the curve along one edge of the surface differs from
the shape along the opposite edge.
Looking first at the polynomial model, since the polynomial modeler cannot
model interaction, the opposite edges of the model surface will have the same
shape. The ANN model is shown in Figure 6.8. Compare the edge running from
the origin out along the Z axis to the opposite edge. The edge running from the
origin (sqFeet
¼
500) along the Z axis shows a lot size contribution that
increases, then levels off. Along the opposite edge (sqFeet
¼
5,000) the
contribution of lot size is a steady consistent increase. Clearly the contribution
of lot size to price interacts with home size (sqFeet).
Issues in Model Interpretation
In reviewing theANNmodel plot, it is important to remember that we are seeing a
plot of the model surface depicting how variations in lot and sqFeet affect price.
All other inputs are held constant. The values of these other inputs are represented
