Chapter Evolving Connectionist and Fuzzy Connectionist Systems: Theory and Applications for Adaptive, On-line Intelligent Systems

9.2. EFuNN-based intelligent agents for adaptive, on-line prediction of the NZ
SE40 stock index
Here an intelligent agent that learns and predicts in an on-line, adaptive mode the
SE40 data is realised as an EFuNN [31]. The same input and output variables are
used as in the experiment with the NZSE40 data in section 3. The following
evolving parameter values are used: sensitivity threshold
Sthr=0.92, error
threshold Errthr=0.05, number of rule nodes rn=910; after pruning this number is
730; learning rate lr=0. The SE40 daily change is predicted on-line based on the
evolving of the EFuNN on the previous data. The root mean square error is
RMSE= 0.22 (on the last 49 test data points) while the random walk gives
RMSE=4.32.
Fig.10. An EFuNN evolved as an intelligent agent and tested incrementally
on the NZ
SE40 difference data
Fig.11. The total activation of the rule nodes of the EFuNN evolved as an intelligent agent
from the NZ SE40 data
0
200
400
600
800
1000
1200
1400
1600
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0
100
200
300
400
500
600
700
800
900
1000
0
100
200
300
400
500
600
700

---

137
Fore more EFuNNs were evolved to predict two, three, four and five days ahead.
The test error RMSE for them is correspondingly 0.25, 0.28, 0.45, 1.26, 2.78. It
can be seen that even 5 days prediction will give a better result than the random
walk one-day prediction. That justifies the use of EFuNNs for this particular task.
As EFuNNs have principally the same structure as 
FuNNs, fuzzy rules can be
extracted as explained in section 3. Fig.11 shows the total activation of the rule
(case) nodes of the evolved EfuNN before pruning.