RBF ANN is also considered of multilayer and forward
feeding configuration. It has the ability to simplify
complex problems, by means of non-linear mapping.
As the complexity increases, adding more perceptron
layers to ANN could lead to longer computational
time and hence slower convergence. Moreover,
discrimination lines are unable to separate among
clusters that are naturally. In these scenarios however,
it may be more tangible to incorporate into
ANN a circular shape discriminator, e.g., Gaussian,
whose extent covers most of the plausible region in
each cluster. This configuration consists of 3 layers,
i.e., n-neural input, one hidden, and n-neural output
layers. The connectivity between input and hidden
layers are hypothetical, while the one between hidden
and out-put layers is specified with a set of adaptive
weights that are adjusted during the neural training
process.