Discovery of Positive and Negative Rules from Medical Databases Based on Rough Sets - Advanced Techniques in Knowledge Discovery and Data Mining - page 247

9. Discovery of Positive and Negative Rules 249

This rule is very similar to the following classification rule for disease of
cervical spine:
[Jolt Headache = no]
∧([Tenderness of M0 = yes] ∨ [Tenderness of M1 = yes]
∨[Tenderness of M2 = yes])
∧([Tenderness of B1 = yes] ∨ [Tenderness of B2 = yes]
∨[Tenderness of B3 = yes]
∨[Tenderness of C1 = yes] ∨ [Tenderness of C2 = yes]
∨ [Tenderness of C3 = yes] ∨ [Tenderness of C4 = yes])

disease of cervical spine
The differences between these two rules are attribute-value pairs, from ten-
derness of B1 to C4. Thus, these two rules can be simplified into the following
form:

→

a 1 ∧ A 2 ∧¬ A 3 → muscle contraction headache,

a 1 ∧ A 2 ∧ A 3 → disease of cervical spine.

The first two terms and the third one represent different reasoning. The
first and second terms a 1 and A 2 are used to differentiate muscle contraction
headache and disease of cervical spine from other diseases. The third term
A 3 is used to make a differential diagnosis between these two diseases. Thus,
medical experts first select several diagnostic candidates, which are similar
to each other, from many diseases and then make a final diagnosis from those
candidates. This problem has been partially solved; Tsumoto introduced a
new approach for inducing these rules in [9.13], as induction of hierarchical
decision rules. In that paper, the characteristics of experts' rules are closely
examined and a new approach to extract plausible rules is introduced, which
consists of the following three procedures. First, the characterization of de-
cision attributes (given classes) is done from databases and the classes are
classified into several groups with respect to the characterization. Then two
kinds of subrules, characterization rules for each group and discrimination
rules for each class in the group, are induced. Finally, those two parts are
integrated into one rule for each decision attribute. The proposed method
was evaluated on a medical database, the experimental results of which show
that induced rules correctly represent experts' decision processes.
This observation also suggests that medical experts implicitly look at the
relation between rules for different concepts. Future work should discover the
relations between induced rules.

9.8.2 Relations Between Rules

In [9.14], Tsumoto focuses on the characteristics of medical reasoning(focusing
mechanism) and introduces three kinds of rules, positive rules, exclusive rules
and total covering rules, as a model of medical reasoning, which is an extended
formalization of rules defined in [9.12].