Discovery of Positive and Negative Rules from Medical Databases Based on Rough Sets - Advanced Techniques in Knowledge Discovery and Data Mining - page 242

244 Shusaku Tsumoto

9.6 What Is Discovered?

9.6.1 Interesting Rules Are Very Few

One of the most important observations is that there were very few rules
interesting or unexpected to medical experts compared to the number of
rules extracted from the data sets.
Table 9.4 shows the mentioned results. The second column denotes the
number of positive and negative rules obtained from each data set. The third
column denotes the number of positive and negative rules interesting or unex-
pected for domain experts. For example, the first row shows that the number
of induced positive rules in headache is 24,335, but the number of interesting
rules are 114, which shows that only 0.47% of rules are interesting for domain
experts.
This table shows that the number of interesting rules are very few: even
in the case of meningitis, only 6.5% are interesting, which suggests that the
interpretation part of domain experts is very hard for the knowledge discovery
process.
Next, we show several examples of rules that are interesting to domain
experts.

9.6.2 Positive Rules in Differential Diagnosis of Headache

In the domain of differential diagnosis of headache, the following interesting
positive rules were found.
[Age < 20] ∧ [History : paroxysmal] → Common Migraine

(Coverage :0.75)
This rule is said to be interesting if it is compared with the following rule:
[Age > 40]

∧ [History : paroxysmal] → Classic Migraine

(Coverage :0.72)
These two rules include two parts; although the values of the attribute
“Age” are different, those of the attribute “History” are the same. This sug-
gests that the attribute “Age” is important for differential diagnosis between
common migraine and classic migraine.

Table 9.4. Number of extracted rules and interesting rules.
Induced positive rules Interesting rules

Headache 24,335 114 (0.47%)

CVD 14,715 106 (0.72%)

Meningitis 1,922 40 (2%)

Induced negative rules Interesting rules

Headache 12,113 120 (0.99%)

CVD 7,231 155 (2.1%)

Meninigitis 77 5 (6.5%)