a set of rules and tree nodes, perhaps on the order of tens. Each rule
often consists of fewer than 10 predicates. In contrast, consider an asso-
ciation rules model that may contain tens of thousands of rules, each
consisting of typically 2 to 5 items and a support and confidence value.
The architecture of the DME can also impact the amount of raw
data storage required. Three architectures are described here:
• In-database DME: The DME resides within the execution
engine of the database. It may be in the same process for
optimal performance and data access. This does not include
DMEs that are invoked only through a common SQL
interface while the DME resides in a separate process space;
this requires data to be transferred from the database engine
to that separate DME process.
• Independent-server DME with direct data access: The
DME is not hosted in the database, but can access database
data directly stored in the database, either through
proprietary data access interfaces or standard interfaces
such as JDBC or ODBC. Such environments do not require
• Independent-server DME with staging data: The DME is
not hosted in the database and requires temporary storage
of data, possibly in a proprietary format or separate data
marts. Some businesses may require data to be staged
separately from operational data stores or the corporate
data warehouse to avoid performance impacts, uncertain
data storage impacts, or security concerns.
These three architectures are depicted in Figure 15-1.
Data Access Layer
Data mining engine data access architectures:
(a) In-database DME, (b) Independent-server DME with direct
data access, and (c) Independent-server DME with staging data.