Features

Most of the time data mining is done over a single flat table. However, in some situations we need to do data mining over multiple tables, perhaps because the data is too large, or remote, or distributed, or for other reasons.

In previous KDnuggets News I asked

"...which data mining software can do analysis over multiple tables."

SQL server has some of that capability.

David Jensen writes:

One candidate would be Proximity, an open-source environment for knowledge discovery in relational data produced by my research group. Details can be found at:

http://kdl.cs.umass.edu/proximity/index.html

Proximity runs under Linux, MacOS, and Windows.

Louis Duclos-Gosselin writes:
What you talk is probably a data mining suite which includes one kind of data integration. SAS Entreprise Miner with SAS Data Integration Studio with SAS Access do the job perfectly.
Other BI tools can do it also - Oracle, Cognos, Business Objects.

Michael Berthold writes
you are probably referring to real multi-relational data analysis. I am not even sure I know of many algorithms that do that without first flattening the data. Otherwise I would put all of the pipeline architectures into this category since you can easily read in various tables, combine/merge/aggregate them in different ways and run the usual analysis methods. Clementine, Insightfuls iMiner, and - of course - KNIME would fall into that category (among others, no doubt).

Dusan Toman writes
Miner3D does interactive data visualization and data analysis over data from multiple sources. Users can use screen layouts with multiple live charts to interactively analyze data from several sources at once.

The upcoming Miner3D Release 7.1 (June 2008) comes with extensive support for both Oracle OLAP and SQL Server Analysis Services and empowers users with intuitive analysis of multiple OLAP cubes at once.

Arno Knobbe writes
Mining multiple tables is generally known as Multi-Relational Data Mining (MRDM), or sometimes plain Relational Data Mining. The product my company sells, called Safarii, is the only commercial tool avialable that does MRDM in its general form, to the best of our knowledge. SQL Server has some basic multi-table functionality, through something called a nested table, but the algorithms it offers are not adapted to deal with this complexity. From what they told me last year, Oracle have no intention to support MRDM.

Safarii is a state-of-the-art Data Mining environment for analysing large relational databases. It is unique in its wide support for a new data analysis paradigm, called Multi-Relational Data Mining (MRDM). This makes it an ideal tool for knowledge discovery in a wide range of industries, such as retail, enterprise resource planning, banking, e-commerce, bio-informatics and molecular modelling.

As most Data Mining tools, Safarii mines a given database for subgroups with interesting or surprising characteristics (customers with a high response rate to a mailing, website visitors with above-average spending, a class of chemical compounds with increased carcinogenic potential). Whereas traditional tools only deal with a single table and the flat features stored in there (age, balance, etc.), Safarii is able to involve structural information stored in several tables. This enables Safarii not only to deal with objects that are inherently structural (molecules, weblogs, etc.) and hard to process by traditional tools, but also to integrate data originating from multiple sources such as different business applications, as well as information from previous analyses and domain knowledge. Safarii supports a range of techniques for exploratory data analysis, building predictive models, and deploying the results in an operational environment.

More information concerning Safarii can be found at www.kiminkii.com/safarii.html.