One of the new features in Oracle 9i was a
Oracle Data Mining, a data mining engine which allowed
data analysts and application developers perform a range
of data mining algorithms on data held in the Oracle
database. Oracle 9i came with a
number of mining algorithms, such as Adaptive Bayes
Networks, Clustering and Association Rules, together with
a Java API to allow ODM functions to be included in Java
applications.
Whilst this was useful for Java
programmers, it wasn't all that relevant for PL/SQL
programmers and to remedy this, Oracle Database 10g comes
with a new package called
DBMS_DATA_MINING that provides PL/SQL access to the
data mining engine.
Like the Java API, DBMS_DATA_MINING allows you to build
a data mining model, test it and then apply the model to
provide scores or predictive information for an
application. One of the key differentiators for Oracle
Data Mining is that mining models can be applied directly
to data in the database - there's no need to extract the
data and then separately load it into the mining engine,
meaning that data mining can be now carried out in 'real
time'. The Oracle Data Mining engine can be pointed at any
schema in the database, and if the data needs processing
beforehand (to place continuous and discrete values into
range 'bins') there's also a new accompanying package,
DBMS_DATA_MINING_TRANSFORM to carry this out
automatically.
In addition to the new PL/SQL interface, Oracle Data
Mining in Oracle Database 10g has the following additional
enhancements;
- Features extraction using Non-Negative Matrix
Factorization Algorithm
- Enhanced Data Preprocessing
- Enhanced Adaptive Bayes Network
- Extension of standard Oracle database security to
Oracle Data Mining user data and mining results
- DM4J, the add-in to JDeveloper that allows the
graphical building of mining models
- Support for Support Vector Machines
More information on these new features can be found at;
