Information Technology Reference
In-Depth Information
where
d
0
-
d
9
represent, respectively, (
t
-10) - (
t
-1). This model has an R-
square equal to 0.94974095 in the training set of 80 data points. And as you
can see in Figures 7.2 and 7.3, the model evolved by GEP is an extremely
good predictor, with an R-square of 0.8785305643 on the testing set. Note,
in Figure 7.3, how the most accurate predictions are the most immediate: the
more one ventures into the future the less accurate they become.
160
140
Model
Target
120
100
80
60
40
20
0
-20
0
10
20
30
40
50
60
70
80
Time
Figure 7.2.
Comparing the model (7.13) designed by gene expression
programming with the target sunspots series on the training data.
120
100
Model
Target
80
60
40
20
0
81
82
83
84
85
86
87
88
89
90
Time
Figure 7.3.
Comparing the mode (7.13) designed by gene expression pro-
gramming with the target sunspots series on the testing data.
The remarkable thing about these time series prediction models (the model
(7.13) above and the models (7.9) - (7.12) of the previous section) is that
they are all composed of simple terms involving a quotient. This is some-
thing the algorithm discovered on its own without any kind of prompting on
