Information Technology Reference
In-Depth Information
(2) to induce a predictive model from the data and compare the induced model with
the model used in CDSS. The evaluation of the KB rules against the data warehouse
of diagnosed cases provides the measurements for the validity of the KB in terms of
its accuracy (sensitivity and specificity). The comparison of the existing KB rules
and the induced rules provides a modification (refinement) mechanism for the rules.
Data Mining
Based on the data from the data warehouse, data mining (machine learning) tech-
niques can be used to automatically induce descriptive and predictive models. In
our research, we have used artificial neural networks (ANN), classification rule in-
duction, and decision tree induction. The ANN methods were used in the CDSS
system for the evaluation of clinical depression. The ANN model has been used as
a part of a hybrid expert system. In case of the CDSS for the diagnosis of OSA, the
clinicians require that the induced models should be comprehensible by humans;
therefore, we have utilized the rule induction and the decision tree induction meth-
ods. In our study, we compared the rules in the existing KB with the automatically
induced rules from the data warehouse. Three scenarios for the induced rules are
possible: (1) they confirm the existing KB rules, (2) they contradict the existing KB
rules (provide contradictory examples), or (3) they identify new insights (contribute
new rules to KB).
To demonstrate how the data mining techniques can be used for the modification
of KB rules, we present two examples from our studies.
Example 1
The following KB rule, R3, specifies low OSA risks for young female patients with
normal weight. Normal weight is represented by the Body Mass Index (BMI) less
than 25.
R3: IF BMI
<
25 AND age
=
young AND gender
=
female
9)
R3 is partially contradicted by the rule induced from data, I1, which additionally
includes hypertension (HTN).
THEN OSA
=
no (CF
=
0
.
I1: IF BMI
<
26 AND age
=
young AND gender
=
female
AND HTN
=
yes THEN OSA
=
yes
Since hypertension is one of the important predictors of OSA, Rule 3 could be mod-
ified and it could include HTN
=
no.
Example 2
The following two rules, I2 and I3, were derived from an induced decision tree:
I2: IF BMI
≤
26
.
8 AND HTN
=
no AND gender
=
female
AND age
≤
56 THEN OSA
=
no
