JDM Strategy - Java Data Mining: Strategy, Standard, and Practice - page 105

data miners can obtain desired results. By distinguishing mining
functions from mining algorithms, JDM simplifies the mining pro-
cess for novices, while allowing experts complete control. Mining
functions allow users to specify what they want to accomplish, e.g.,
classify these cases into two categories, or cluster these cases into n
clusters. Novice data miners who are not familiar with the details
of data mining can rely on the vendor to choose a specific algo-
rithm. In addition, many of the settings that control the mining
function or algorithm behavior allow a system determined setting
that instructs the data mining engine to make an intelligent selec-
tion, perhaps based on the nature of the data presented. JDM also
allows for automated data preparation to further support the nov-
ice user. On the other hand, expert data miners, familiar with the
details of data mining, can select the specific algorithm or specify
detailed settings. Expert data miners can prepare their data explic-
itly and instruct the data mining engine not to further transform
their data.
Strategic Objective 7: Recognize conformance limitations
for vendor implementations
Like the field of data mining itself, data mining software vendors
support a wide range of data mining techniques. In some cases,
vendors may specialize in a particular technique like neural
networks, or in a small set of techniques such as classification and
regression, providing a variety of algorithms. As such, it is
important to create a standard that is flexible enough to allow ven-
dors with different focuses to implement and comply with the
standard.
Tactically, we reduce barriers to entry for vendors to support the
standard by making conformance a la cart ; that is, the standard is
naturally segmented into logical units, which can be easily
supported independent of others. Simply, this means that JDM is
organized into packages specific to different types of functionality
that vendors can optionally choose to support. Of course, there is a
relatively small core set of packages that must be supported, yet
vendors can select to implement a narrow subset of functionality,
for example, only the classification mining function with the deci-
sion tree algorithm. Other vendors may choose to implement only
interfaces at the mining function level, leaving out any algorithm
details. This flexibility makes the JDM standard much more accessi-
ble to niche vendors.