EVOLUTION OF NEURAL NETWORK AND POLYNOMIAL NETWORK - Knowledge Mining Using Intelligent Agents - page 79

Table 3.5. Architectures evolved for different sets of data.

Datasets Number of Number of neurons in hidden layers

hidden layers

Layer 1 Layer 2 Layer 3

AUSTRA
Set1 1 5 — —

Set2 1 10 — —

BALANCE
Set1 0 — — —

Set2 0 — — —

CREDIT
Set1 1 3 — —

Set2 1 1 — —

HABERMAN
Set1 0 — — —

Set2 0 — — —

HOUSE
Set1 1 2 — —

Set2 0 — — —

IRIS
Set1 0 — — —

Set2 0 — — —

MONK
Set1 0 — — —

Set2 1 10 — —

PIMA
Set1 1 10 — —

Set2 1 10 — —

WBC
Set1 0 — — —

Set2 1 5 — —

WINE
Set1 3 10 10 3

Set2 0 — — —

self-organizing network. In each layer, a set of qualified PDs is retained
based on magnitude of modeling error. The outputs of these retained PDs
are manipulated in the subsequent layer PDs for further modeling error
minimization. New layers are introduced till the modeling error reaches
some pre set level.