Information Technology Reference
In-Depth Information
phase consists of an expert revision for the proposed solution, and finally, the retain
phase allows the system to learn from the experiences obtained in the three previous
phases, consequently updating the cases memory.
A key element in a CBR system is a case, which can be defined as a past experi-
ence [10] and is composed of three elements: problem description, problem solution,
and the final state obtained after applying the solution. A case in the system presented
in this work contains information related to the patient, the rules, the proposed classi-
fication, and the probes marked as irrelevant or important. The case is defined by the
following expression:
p
r
i
=
(
id
,
S
=
(
A
,...,
A
),
C
,
C
)
j
1
n
i j
I
I
=
{ 1
i
,...,
i
}
is the set of individuals/cases, A is the set of all
where
and
s
p
r
A represents the probe i ,
C is the predicted class and
C the actual
the probes,
class.
In addition to the cases memory, our system incorporates a memory of rules that
contains the information extracted through the knowledge extraction techniques. The
memory of rules is structured as follows:
}
R
=
{ 1
r
,...,
r
r
=
( 1
l
...
l
)
c
,
with
where
i
m
j
l
=
(
d
,
o
,
)
/
d
D
,
o
O
S Irr
A
where R is the set of rules from the decision rules,
,
s
ts
s
ts
t
s
l contains a set of discretized
D is the discretization value for the probe
A ,
probes, an operator and a real value,
O
=
{
=
,
,
>
,
<
,
,
}
Ir S is the set of probes marked as irrelevant,
operator, and
p
c
j C
.
When a new case is classified, a new decision rules are generated in the revise
stage. A set of rules are extracted which provide knowledge about the relevance of the
probes in the clustering and classification process. Figure 1 shows a scheme of the
techniques applied in the different stages of the CBR cycle. As seen in Figure 1, the
important probes that allow the classification of patients are recovered in the Retrieve
phase. The Retrieve phase is divided into 6 sub-phases: pre-processing through RMA,
removal of irrelevant variables, uniform distribution, probes without meaningful cut-
off points, and correlated variables. In the Reuse phase the patients are grouped by
means of an ESOINN neural network. Then, the patients with no prior classification
are assigned to a group using the nearest cluster. In the Revise phase the RIPPER [43]
algorithm is applied for extracting knowledge about the most important probes for the
classification, and the MDS technique [18] [19] [20] is used to make a representation
in low dimensionality. Finally, in the Retain phase, the knowledge is updated. This
knowledge includes the case classification, the decision rules obtained, and the infor-
mation associated with the importance or irrelevance of certain probes extracted from
the rules. Figure 1 shows the scheme of the CBR system proposed within this study.
Search WWH ::




Custom Search