Information Technology Reference
In-Depth Information
Also important is the fact that the best-of-experiment solution created with
decision trees classifies correctly 149 out of 150 instances, a feat only achieved
by the three-step approach of the basic GEA (see Decomposing a Three-
class Problem in section 4.2.3) and the multicellular system with multiple
outputs (see section 6.6 Multiple Cells for Multiple Outputs: The Iris Prob-
lem). Its structure is shown below:
0123456789012345678901234567890
PaQPPbTTQaccbbcaacaab5218839495
C = {5.53, 1.69, 6.89, 9.67, 3.12, 2.53, 4.24, 3.96, 4.99, 1.79} (9.6)
As you can see in Figure 9.16, it encodes a relatively compact decision tree
with 15 nodes, with the PETAL_LENGTH (“P”) at the root. It is also inter-
esting to note that the attribute SEPAL_LENGTH (“S”) is not used by this
extremely accurate model, which was somewhat of a surprise since in all the
previous best-of-experiment models analyzed thus far (see models (4.14)
and (4.20), and the model (6.10) expressed in Figure 6.13) all the attributes
were used in the trees. In other cases, though, it is the SEPAL_WIDTH (“T”)
that is not used, which most probably means that both these attributes are
interchangeable.
It is also interesting to compare the structures of this extremely accurate
decision tree with other DTs equally accurate in order to see, among other
things, how these trees grow. For instance, the 30 decision trees presented
below all have fitness equal to 149 and all of them are extremely parsimoni-
ous with just 13 nodes each:
QPSQccQabScbccaaaacbb4372201889-[1]
C = {9, 1.77, 0.68, 4.96, 1.59, 1.02, 6.99, 5.74, 1.59, 6.79}
SQQPTPcabcbbcccaccbbc5289076015-[2]
C = {3.12, 1.25, 1.62, 9.8, 1.25, 6.02, 7.88, 5.03, 1.77, 2.69}
PQQPTcQabcbbccaaaaacb6775813381-[3]
C = {3.64, 1.79, 3.64, 1.72, 0.56, 2.79, 4.9, 1.57, 3.19, 6.53}
PQPPTTcabcbcbabccbcba5740169053-[4]
C = {2.01, 3.1, 2.05, 8.27, 5, 4.97, 2.72, 1.64, 8.82, 4.89}
PQcaPTSQbcbbcbcccbcaa3520492834-[5]
C = {7.45, 4.58, 6.49, 5.04, 4.93, 0.64, 3.33, 4.44, 1.51, 3.09}
PQQQTcQabcbbcaacbbbac2008165834-[6]
C = {1.68, 3.11, 4.92, 0.67, 0.87, 1.76, 8.79, 6.8, 0.62, 8.19}
