Information Technology Reference
In-Depth Information
As you can see in Table 9.5, for this problem we are going to use small
population sizes of just 30 individuals evolving for 1000 generations (ex-
actly the same values used in the experiments summarized in Tables 5.9 and
6.7). However, as you can see by their performance, decision trees need more
time than all the other GEP systems to design good solutions to this particu-
lar problem (average best-of-run number of hits of 337.40 versus 339.34
obtained with GEA-B; 339.05 with GEP-NC; 339.13 with GEP-RNC; 339.19
with MCS; and 339.06 with MCS-RNC), but they get there nonetheless. In
fact, the best-of-experiment solution is an exceptional DT model with a fit-
ness of 342 in the training set and 170 in the testing set, which corresponds
to a training set accuracy of 97.714% and a testing set accuracy of 97.701%
and, therefore, is as good a model as the best models evolved with the other
Table 9.5
Performance and settings used in the breast cancer problem.
Number of runs
100
Number of generations
1000
Population size
30
Number of fitness cases
350
Attribute set
A-I
Terminal set / Classes
a b
Random constants array length
10
Random constants type
Rational
Random constants range
[0, 1]
Head length
25
Gene length
76
Mutation rate
0.044
Inversion rate
0.1
IS transposition rate
0.1
RIS transposition rate
0.1
One-point recombination rate
0.3
Two-point recombination rate
0.3
Dc-specific mutation rate
0.044
Dc-specific inversion rate
0.1
Dc-specific transposition rate
0.1
Random constants mutation rate
0.01
Fitness function
Eq. (3.8)
Average best-of-run fitness
337.40
