Design of the JDM API - Java Data Mining: Strategy, Standard, and Practice - page 169

javax.datamining
MiningObject

Data Specification Objects Settings Objects

javax.datamining.data
PhysicalDataSet

javax.datamining.base
Model

javax.datamining.base
BuildSettings

javax.datamining.base
Task

javax.datamining.data
LogicalData

javax.datamining.task.apply
ApplySettings

javax.datamining.data
Taxonomy

javax.datamining.supervised
TestMetrics

javax.datamining.supervised.classification
CostMatrix

Figure 8-1 JDM mining objects.

8.1.1 Data Specification Objects

Data specification objects are used to describe the input data for
mining. It is important for the DME to know data attribute charac-
teristics so data can be mined effectively. For example, in the CUS-
TOMERS dataset, as discussed in Section 7.1.3, the attribute Marital
Status might have numerical values such as 1, 2, 3, 4 to represent
unmarried, married, divorced, and widowed, respectively. Here, the
DME needs to know this is a categorical attribute as opposed to a
numerical one.
JDM defines physical and logical data objects to encapsulate
input data details common across all mining functions. The Physical-
DataSet object captures the physical data characteristics, such as data
location, physical attribute names, data types, and what role the
attributes play. The LogicalData object specifies how to interpret the
data. It captures logical data characteristics, such as whether an
attribute is categorical/numerical/ordinal, prepared/unprepared,
and what are the valid values. Separating logical data from physical
data allows applications to manage them independently and map
multiple PhysicalDataSet objects to a single LogicalData object and
vice versa, thereby allowing reuse of both physical and logical spec-
ifications across mining operations and activities. For example, in a
production system, the physical data may change weekly such that
each dataset has a different table name, yet the way to interpret
those attributes logically remains the same. In this case, the data
miner maps multiple physical dataset objects with one logical data