Databases Reference
In-Depth Information
Drag the sliders for SepalLength and SepalWidth back to their original
positions at 6.1 and 3.2 respectively.
In the options panel, change the “Classification” value from Setosa to
Versicolor and then to Virginica. How do the likelihoods compare?
Change the axis variables. Put SepalLength on the X axis and SepalWidth
on the Z axis. How much do these two attributes contribute to the
likelihood of Setosa?
To better grasp contributions of PetalLength and PetalWidth in classifying
Variety independent of the other attributes, it is best to build a model using only
those two attributes as input.
Create a derived dataset containing only PetalLength, PetalWidth, and
Variety.
Drag the SVM Classifier down to the newly derived dataset to build
another model.
View the confusion matrix for this model. How does it compare with the
model that used all four input attributes?
View the model in the Classification Surface viewer side-by-side with the
original four input SVM model.
The resulting surfaces for Setosa, Versicolor, and Virginica are in
Figure 5.6. The likelihood of Setosa is greatest when both PetalLength and
PetalWidth are low. The likelihood of Versicolor is greatest when both are
in the mid-range. They drop to much lower values when only one of the two is
in that mid-range. The likelihood of Virginica is greatest when the values of
both are high.
Figure 5.6
Likelihood Plots - Setosa, Versicolor and Virginica
