Databases Reference
In-Depth Information
We need to acknowledge that some work has been done with regard to the
study of
interestingness
of discovered patterns in the context of association
rules mining (for example [38]). Yet, even within this particular area, it has
been fairly noticed in [5] that there is no consensus on how the interestingness
of discovered patterns should be measured, and that most of DM research
avoids this thorny way reducing
interestingness
to
accuracy
and
comprehen-
sibility
.
Disregarding the relevance issues, DM frameworks ignore also the issues
of DM artifact development and DM artifact use. Here and in the following
text by DM
artifact
we mean either “hard/technical” artifacts like DM model,
DM technique or its instantiation, collection of DM techniques that are part
of DM system or DM embedded solution, or “soft/social” artifacts like some
organizational, operational, ethical and methodological rules that focus on
different considerations of risks, costs, etc.
In Sect. 3 we first refer to the traditional information system (IS) frame-
work presented in [9] that is widely known in the IS community and is a syn-
thesis of many other frameworks considered before it. This framework takes
into account both the use and development aspects beside the technical ones
in the IS area. Further we consider more detailed IS frameworks from the use
and development perspectives.
In Sect. 4 we introduce our sketch for the new framework for DM research
based on the material included in Sects. 2 and 3. We strongly emphasize the
relevance aspect of DM research, trying not to neglect the rigor. This means
that beside the technological aspects also the organizational and human as-
pects should be equally taken into account. Thus, our framework for DM
research suggests a new turning point for the whole DM research area.
We conclude briefly in Sect. 5 with a short summary and further research
topics.
Some materials presented in this chapter are the results of our earlier
work [34-36].
2 Review of Some Existing Theoretical Frameworks
for DM
It this section we review basic existing foundations-oriented frameworks
for DM based on different paradigms, originating from statistics, machine
learning, databases, philosophy of science, and granular computing and the
most well-known process-oriented frameworks, including Fayyad [13], CRISP-
DM [6], and Reinartz's [37] frameworks. We present our conclusions for these
groups of DM frameworks and then analyze the state of art in DM research
in general.
