Trends in Data Mining and Knowledge Discovery - Advanced Techniques in Knowledge Discovery and Data Mining - page 16

z ordered - attribute with discrete values that are ordered.
z cyclical - attribute with discrete values that are ordered and cyclical, e.g.,
weekdays.
z sequence time - attribute containing time measurement units.
z sequence - attribute containing the sorting key of the related attributes.
The OLE DB-DM supports the following DM models:
z classification when the predicted attribute is categorical.
z regression when the predicted attribute is continuous.
z clustering.
z association (data summarization) including association rules.
z sequence and deviation analysis.
z dependency modeling used to identify dependencies among attributes.
Two main advantages of the OLE DB-DM are:
z it can interface with PMML, because all of the structure and content of a
DMM may be expressed as an XML string in PMML format
z it can interface with the OLAP technology.
The technologies described above can be used to integrate and semiautomate
the DMKD process, on the level of manipulation and sharing data, and on the
level of data models. XML-based technologies can be used to store data and DM
data models and to provide communication protocols between DM tools. OLAP
can be used during the data preprocessing step, and OLE DB-DM can be used to
integrate DM tools with relational DBMSs.
1.4 Future of Data Mining and Knowledge Discovery?
IDC, a well-known provider of technology intelligence and industry analysis,
estimates that the data mining tools market will reach $1.85 billion in 2006. In
1998, Simoudis of IBM predicted that “within five years, [data mining] will be as
important to running a business as the business systems are today;” his prediction
has proven to be already correct (2004). On the other hand, many business
managers are willing to conduct DMKD on their data but they are not sure where
to start [8].
The DMKD community developed several successful DM methods over the
last few years. A survey of software implementations of DM methods presents a
comparison of 43 existing implementations of DM methods [35]. Unfortunately,
just having a variety of DM methods does not solve the problems of DMKD, like
the necessity of integrating DM methods, integrating them with the DBMS, and
providing support for novice users.
To provide a framework to address these issues we start by defining DM
methods and DM tools. A DM method is simply an implementation of a DM
algorithm; a DM tool is a DM method that can communicate and operate in the
DMKD environment. Development of DM tools or upgrading the existing