Decision Tree Induction - Gene Expression Programming - page 369

classifies correctly all the testing instances, which corresponds to a training
accuracy of 99.0% and a testing set accuracy of 100%:

0123456789012345678901234567890
PQQPTcQabcbbccccbcbcc9336038121

C = {3.16, 2.61, 1.76, 1.61, 3.11, 5.64, 2.25, 1.58, 1.74, 4.91} (9.7)

As you can see by its expression in Figure 9.18, it encodes a very compact
decision tree with just 13 nodes with the PETAL_LENGTH (“P”) at the
root. Note again that this time it was the attribute SEPAL_LENGTH (“S”)
that was not used to distinguish between the three kinds of iris plants.
Let's now see how the EDT-RNC algorithm deals with complex problems
with mixed attributes.

a.

0123456789012345678901234567890
PQQPTcQabcbbccccbcbcc9336038121

C = {3.16, 2.61, 1.76, 1.61, 3.11, 5.64, 2.25, 1.58, 1.74, 4.91}

b.

PETAL_LEN

d >

4.91

4.91

PETAL_WID PETAL_WID

d > d > 1.61 1.61 1.61 1.61

PETAL_LEN SEPAL_WID Vir PETAL_WID

d > d > d

2.25 2.25 3.16 3.16 1.74

>

1.74

Set Ver Vir Ver Ver Vir

Figure 9.18. Testing the generalizing capabilities of the EDT-RNC algorithm on the
iris data. a) The linear representation of the model. b) The decision tree model. This
model has 99.0% accuracy on the training set and generalizes outstandingly well
on the testing set, with 100% accuracy.