Trends in Data Mining and Knowledge Discovery - Advanced Techniques in Knowledge Discovery and Data Mining - page 20

DM tool called Oracle Darwin ® , which is a part of the Oracle Data Mining Suite
[52]. It supports DM algorithms like neural networks, classification and regression
trees, memory-based reasoning (based on k -nearest neighbor approach), and
clustering (based on k -means algorithm) [13], [17]. Their solution integrates with
the Oracle 9i DBMS.
These products provide tools to automate several steps of the DMKD process
like preparation of the data and DM. However, they only partially solve the issue
of semiautomation of the entire DMKD process because they do not provide an
overall framework for carrying out the DMKD process.
1.5 Conclusions
Currently the DMKD “industry” is quite fragmented. It consists of research groups
and field experts that do not work closely with decision makers. This is caused by
a situation where the DMKD community generates new solutions that are not
widely accessible to a broader audience and are difficult to use. Because of that,
and the high cost of performing the DMKD process, DMKD projects are
undertaken by companies which can afford them and urgently need to analyze
large amounts of data they constantly collect, but many other businesses reject it
because of the costs involved. To solve this problem we need to semiautomate the
DMKD process, provide integrated DM tools and services, thus making the
DMKD process easier and less expensive to use by the end user.
The technologies described in this chapter (XML, XMP-RPC, SOAP, PMML,
UDDI, OLAP, and OLE DB-DM) will play a significant role in the design of the
such next-generation DMKD process framework. These technologies will make it
possible to build DM toolboxes that span multiple DM tools; to build knowledge
repositories; to communicate and interact between DM tools, DBMSs, and
knowledge repositories; and most importantly, to semiautomate the entire DMKD
process. These technologies also can be used to deploy the DMKD process that
will include elements that run on different platforms because they are platform-
independent. Another advantage of these technologies is that they will bring the
DMKD industry to a new level of usability. New users, who will follow these
standards, in spite of their lack of deep knowledge of DMKD, will be exposed to
and attracted by DMKD applications.
In addition to the design and implementation of a new DMKD framework, a
more traditional course of action will need to be undertaken [37]. It includes
design and implementation of a new generation of high-performance DM systems
that incorporate multiple DM methods [76] and are capable of mining
heterogeneous sources of knowledge like multimedia data [77], can visualize the
results, and can handle huge amounts of complex data. One of the goals in
designing such systems should be the design of better user interfaces. This will
result in a wider acceptance of the products, particularly by midsize and small
companies with limited technical skills. Another important issue is to learn about
the user perception of the novelty, understandability, and simplicity of the
knowledge generated by the DMKD process. We also should take into account the