Databases Reference
In-Depth Information
Tabl e 4 .
Time needed to extract action rules by
DEAR-2
DataSet
Action Rules DEAR 2
Breast Cancer
3
s
Cleveland
54
min
20
s
Hepatitis
51
min
53
s
Fig. 3.
DEAR
and
DEAR2
interface
attribute before any action rule is constructed. The next two tables show the
time needed by
DEAR
and
DEAR-2
to extract rules and next action rules
from three datasets:
Breast Cancer
,
Cleveland
,and
Hepatitis
. These three
UCI
datasets are available at [
http://www.sgi.com/tech/mlc/db/
]. The first
one has 191 records described by ten attributes. Only
Age
is the stable at-
tribute. The second one has 303 records described by 15 attributes. Only two
attributes
age
and
sex
are stable. The last one has 155 records described by
19 attributes. Again, only two attributes
age
and
sex
are stable.
The interface to both systems,
DEAR
and
DEAR-2
, is written in Visual
Basic. The first picture in Fig. 3 shows part of the interface to both systems.
The user has an option to generate the coverings (see [7, 8]) for the decision
attribute and next use them in the process of action rules extraction or, if
he prefers, he can directly proceed to the rules extraction step. It is recom-
mended, by
DEAR-2
, to generate the coverings for the decision attribute if
the information system has many classification attributes. By doing this we
usually speed up the process of action rules extraction. The second picture in
Fig. 3 shows how the results are displayed by
DEAR-2
system.
5 Conclusion
Generally speaking, actionable knowledge discovery based on action rule min-
ing can provide a coarse framework for users in applying the rules to objects.
We see that there is a clear need for an effective representation of actionable
knowledge by giving the users exactly the information they need. Hence, we
