companies and some smaller ones. OMG maintains a board of
directors. Participation is open to all companies; each member
company receives one vote to ensure all participants have an effec-
tive voice in the standards-setting process. To initiate an OMG stan-
dard-setting activity, members must submit a Request for Proposal.
Standards are approved by member voting.
The OMG's CWM for data mining formed an expert group
comprised of representatives from IBM, Oracle, and Hyperion. Here,
the goal was to define a complete specification of the syntax and
semantics needed to export and import shared warehouse metadata
and the common warehouse metamodel [CWMI 2000]. CWM/DM
defined a set of objects and states to capture data mining metadata
(e.g., objects for models, settings, and attribute usage, among others).
CWM/DM was not intended to capture the details of individual
models, but to serve as a framework that individual vendors could
extend. Unfortunately, the scope and implementation requirements
of CWM itself limited its adoption among some data mining ven-
dors. Since the CWM 1.1 release, no further extensions to CWM/DM
have been made.
SQL/MM Part 6 Data Mining
SQL/MM Part 6 Data Mining (SQL/MM DM) is geared toward SQL
access to data mining in databases. SQL/MM DM took the standards
effort a step further by recognizing the need for a standard SQL-
based interface for building, testing, and applying models, while
leveraging the model representation format available from PMML.
SQL/MM DM does not specify how to represent models but sug-
gests that standards such as PMML are likely candidates for model
representations stored in database tables. SQL/MM DM began with
the most commonly used data mining techniques: classification,
regression, clustering, and association . It also prescribed the use of sep-
arate settings to control the behavior of the underlying algorithms.
SQL/MM DM used SQL user-defined types for the interface defini-
tion. Although this approach may be convenient from an implemen-
tation or user perspective, being object-based as opposed to defining
syntactic extensions to the SQL language itself, the implementation
of user-defined types by database vendors limits the flexibility of
including data mining statements in complex SQL queries, also
called analytical pipelines . Like PMML, SQL/MM DM continued to
evolve, becoming more precise in its definitions, and expanding
functionality to map to progress with PMML, CWM, and JDM.