The 10.2 release of Oracle Data Mining includes the following new algorithms and features:
Decision Tree can be used to build classification models; Decision tree models always have rules. For more information, see Decision Tree.
Oracle Data Miner builds, tests, and applies models using the Java interface. Models built using the ODM 10.1 Java interface are not compatible with models built using the ODM 10.2 Java interface. There is no automatic way to migrate ODM programs written using the ODM 10.1 Java interface to programs that use the ODM 10.2 Java interface. Models built using the ODM 10.2 PL/SQL interface and the ODM 10.2 Java interface are compatible.
Receiver Operating Characteristics (ROC) analysis is a useful method for evaluating classification models. ROC curves provide a means to compare individual models and determine thresholds which yield a high proportion of positive hits.
Data mining models may need to be moved between Oracle databases or schemas. For example, data mining specialists may build and test data mining models on one dedicated system. After the models are built and tested, the models may be moved to another system used by an application. Because the system where the models are developed and the system where the models are deployed usually do not share the same database, the model must be exported from the system where it was developed and then imported to system where it will be used by applications.The ODM programmtic interfaces support moving models.
One-class SVM classifiers can build models when there are no counterexamples. Such models can used for anomaly detection.
Predictive Analytics is based on the PL/SQL package DBMS_PREDICTIVE_ANALYTICS that automates the later stages of data mining; it provides the following functionality:
Copyright © 2005, Oracle. All rights reserved.