Computational Intelligence Techniques for Classification in Microarray Analysis - Computational Intelligence in Healthcare: Advanced Methodologies - page 290

methods extract similar information to classify individuals according to decision
rules. The results are similar for the different methods.
The general objective of extraction of knowledge techniques is to provide a human
expert with information about the system-generated classification by means of a set of
rules that support the decision-making process. It should be noted that knowledge
extraction techniques are not intended to substitute the rationale and experience of a
human expert during a diagnosis, rather to complement the process and serve as an
additional methodology or guideline for common procedures in analysis.
The process is described in the following steps. Let I be the set of individuals and
s the number of probes once the filtering process has finished:

'
1

'

f : A × ... × A → A × ... × A

r 1 s s

'
1

'

( a ,..., a ) → f ( a ,..., a ) = ( a ,..., a )

1 s r 1 s s

'

A

∈ [ 0 ,

is the value of the term i using the function

f where and is obtained in

r

the following way:

a − min( A )

a

'

=

i i

i

max( A ) − min( A )

i i

f

in a series of predefined levels t . Finally the values are discretized by means of

u

s

'
1

'

f : A × ... × A → D × ... × D

u s

'
1

' '

1

' '

1

'

( a ,..., a ) → f ( a ,..., a ) = ( d ,..., d )

s u s s

1

where ∪ 1

D = i ⋅ d

'

= d f

, we can say that if applying the function ,

i j u

t − 1

i ∈ { 0 ,..., t −

'

d j ∈ D d ∈ [ d − 1 /( 2 t ), d + 1 /( 2 t )]

with .

Once the transformation is finished, the set of individuals is determined by the sub-

if

i j j

s

..' of the data, and RIPPER is used to generate the rules that clas-
sify the individuals. The use of RIPPER, allows rules to be obtained for classifying an
individual

I ⊆ D × × D

set

i k ∈ I '

c by means of rules similar to:

to the class

r = ( 1 l ∧ ... ∧ l ) → c

i m j

where d is the value for the attribute p for the individual i . In this way the set is
defined for rules R that classify the individuals for each of the classes.
The input corresponds to the discretization of the values (if the reuse phase has been
successful). Subsequently, knowledge extraction is applied through the RIPPER. Fi-
nally, the relevant information extracted is stored (probes inconsequential, important