Computational Intelligence Techniques for Classification in Microarray Analysis - Computational Intelligence in Healthcare: Advanced Methodologies - page 283

the two previous stages. In this sense, the system presented in this work provides a
knowledge extraction method in the Revise phase. This method analyses the steps
followed in the retrieve and reuse stages, and extracts knowledge which is then for-
malized in rules. In this way, the human expert can easily evaluate the classification
and extract conclusions concerning the efficiency of the classification process. A
RIPPER algorithm is used.

3.4 Retain

If the human expert identifies relevant information at the revise stage, the knowledge
is acquired and the information obtained is stored. The information that is stored cor-
responds to the classifications considered correct, the decision rules generated that are
considered relevant, and the probes marked as irrelevant. The information stored is
divided into the cases memory I and the memory of rules R and Ir S . Figure 1
shows the structure of the retain stage. Taking the revision of the expert into
account, the system learns from the new experience and stores the information that
the expert established as relevant. The stored information can include probes, classifi-
cations and rules.

4 Innovative Computational Intelligence Techniques for
Dimensionality Reduction and Classification Improvement

This chapter details the innovative computational techniques included in the CBR
phases of the system. The innovations consist of dimensionality reduction, classifica-
tion improvements and extraction of knowledge technique. As the computational
intelligence algorithms are included in the different phases of a CBR cycle, in this
section we are going to present each of the novel algorithms as a part of the phases of
the CBR cycle. Figure 1 details the steps followed in each of the stages of the CBR
cycle. The structure of the CBR system proposed will now be explained in detail,
presenting innovative techniques modelled in each of the stages of the CBR.

4.1 Filtering

This computational intelligence technique is carried out in the retrieve phase of the
CBR cycle. The filtering phase is carried out on I together with the new case i .
The filtering is only applied to those probes not associated with any of the rules. First,
a pre-processing of the data is conducted using RMA. Then, the 5 filtering sub-phases
are executed: removal of control probes, removal of erroneous probes, removal of low
variability probes, removal of probes with a uniform distribution, and removal of
correlated probes. These five sub-phases are outlined in the following paragraphs.

4.1.1 RMA
This phase begins once the laboratory experiment with microarrays has been com-
pleted. The researcher obtains various files that contain gross intensity values. Prior to
analyzing the data, it is important to complete the pre-processing phase, which