Computational Intelligence Techniques for Classification in Microarray Analysis - Computational Intelligence in Healthcare: Advanced Methodologies - page 288

3. Once all input patterns have been introduced then a KS-Test [33] is carried out
in order to determine if the density distribution for the neurons in each group
follows a normal distribution. If so, the learning procedure is finished; otherwise
the next pattern is processed. The value of chosen is 0.05.

α

Once the cases have been distributed in the meshes, it is necessary to assign each of
the meshes to a class according to the following procedure: Let

I be the set of indi-

E

G the set of clusters created by
means of the ESOINN neural network, defined as

viduals once the probes have been filtered and

E E E

G = { g / g ⊆ I ' }

, where

E
i

E
j

E E

j

E

g ∩ g = φ , ∀ i ≠ j g , g ∈ G

. Let C be the set of existing classes with

i

c j ∈ C

is the class j in the set. We can say that the mesh for the individuals where

E E

i

E
i j

i ,

i g belongs to a class j ,

g ∈ c

c g when

and is represented as

j

E
i

# I / g # I '

c c

max

j

⋅

j

(8)

# I # I '

c ∈ C

j

c

j

E
i

I ' = { s ∈ I / s ∈ c } I ' / g

c re-

where and is the set of individuals from

c j c

j j

E

i g .
The set of meshes belonging to the class is denoted as

stricted to the group

E

c G and is defined by the

expression (9).

∪ E
E

G

E
c

= g

E
i

(9)

j c

j

g ∈ G

i

j

4.2.2 PAM
The PAM algorithm [16] is executed parallel to the clustering in order to facilitate a
comparison of the results obtained. The classification made by both methods, PAM
and ESOINN, generates an equivalence index between the two methods that deter-
mines the consistency of the reuse phase. The algorithm used for PAM is as follows:

Select the number of clusters depending on C

#

1. .

2. The metric used for the distance is the same as the one used in the ESOINN

network

3. Classify the patients taking all of the variables into account, without any fil-

tering

P P P P

i

P
j

G = { g / g ⊆ I } g ∩ g = φ , ∀ i ≠ j

with

P

G are created, an assignation is made following the pro-
cedure indicated in (8).

4. Once the groups

4.2.3 Equivalence Index
Once the individuals have been classified using both the PAM and the ESOINN neu-
ral networks, the equivalence index for both methods eq is calculated, and the error