Information Technology Reference
In-Depth Information
automated tools. To support verification of a rule-based system, we have constructed
Computer Aided Knowledge Engineering (CAKE) system for the management of
knowledge base.
CAKE system provides automatic detection of the following
problems [7]:
Redundant Rules
Two rules are redundant if
R
i
←
W
i
1
∧ ...∧
W
in
and
R
j
←
W
j
1
∧ ...∧
W
jn
,where
R
}
=
W
j
1
...
Wjn
.
is conclusion,
W
are conditions and
i
=
j
, holds:
{
W
i
1
...
Win
Subsuming Rules
If for two different rules:
R
i
←
W
i
1
∧ ...∧
W
im
and
R
j
←
W
i
1
∧ ...∧
W
in
, holds
{
W
il
∧ ...∧
W
im
}⊆{
W
il
∧ ...∧
W
in
}
, then we say that rule
R
i
subsumes rule
R
j
.
Contradictory Rules
Two rules are contradictory if
R
i
←
W
i
1
∧...∧
W
in
and
¬
R
j
←
W
1
∧...∧
W
n
,where
i
=
j
.
Inconsistent Rules
Two rules are inconsistent if
R
i
←
W
1
∧...∧
W
n
and
R
j
←
W
1
∧...∧
W
n
,where
i
=
j
and
R
i
=
R
j
.
Incompleteness of Rules (Missing Rules)
It is assumed that a knowledge base is complete when the rules include all possi-
ble combinations of the attributes and their allowable values in the rule antecedents
(conditions) and the rule consequents (conclusions). It is important to notice that in
practice most knowledge bases are incomplete, which is a normal situation.
Unused Attributes and Values
System checks if there exist attributes or values which are never used in the rules
in KB.
14.4.4
Validation of KB
Validation of KB is supported by two components:
data warehousing and data
mining.
Data Warehousing
Most of the clinical data are already stored in a digital form, with some clinics using
electronic patient records. The electronic storage of medical data provides a unique
opportunity for a computer-assisted creation, validation, and modification of KB. In
our research, we have used two databases: records of diagnosed depression cases
and a large repository of diagnosed patient records provided by sleep disorders clin-
ics. The data warehouse of diagnosed cases can be used in two ways: (1) to run
the KB rules for each case and compare its results with the clinicians' diagnosis and
