[Ripley, 1996]. The elmNN package employs the extreme
mechanism, i.e., sparse network learning [Huang, et al., 2012;
t al., 2015]. The brnn package employs the Bayesian learning
[MacKay, 1992].
ow how these three packages work, the seeds data was used again.
en neurons were used in three models. The Jackknife test was
all three models. Table 3.8 shows three confusion matrices of
ee models. Figure 3.26(a) shows the ROC curves for these three
All show a similar performance.
The confusion matrices for three MLP models with ten hidden neurons for the
BRNN
NNET
ELMNN
A
B
%
A
B
%
A
B
%
61
10
85.9
62
9
87.3
61
10
85.9
10
55
84.6
7
58
89.2
4
61
93.8
85.9
84.6
84.2
89.9
86.6
87.1
93.8
85.9
88.5
(a) (b)
(a) The ROC curves of nnet, brnn and elmnn models using ten hidden
the seeds data. The figures are the AUC values, which are 0.88, 0.87 and 0.90
nn model, the nnet model and the elmnn model, respectively. (b) The AUC
hree ANN models using three MLP packages with varying hidden neurons for
ata.
performance shown in Figure 3.26(a) was based on the fixed
of hidden neurons. In fact, different numbers of hidden neurons
