166
T.-S. Chon . Y.S. Park' I.-S. Kwak . E.Y. Cha
#I#ti!l'vc,(. 11
11
61
71
81
91
101
11
21
JI
11
.,1
61
,I
111
'11
101
Fig. 8.21. Example of input data and test results in patteming benthic
macroinvertebrates commumtles when traind with the counterpropagation
network. a) and b) actual field data respectively for Genus/Species (X) and Family
(Y), c) trained data for Genus/Species (X), d) new field data for Family (YneJ, e)
predicted data for Genus/Species with new field data (X
U ), and f) actual field data
(X u '). (From Park et al. 2001a).
We further tried to test how the trained network correspondingly responded to
new field data. Fig. 8.21d was another set of field data at the Family level, which
had been collected at the study sites from June to November 1994, and have not
been used for training. The new data set was given to the trained network, serving
as Y new as shown in Fig. 8.20. Fig. 8.21e shows the predicted results at
Genus/Species level (X") after new input data Y new were provided to the trained