MDX or data mining in manufacturing?
Posted by thomasivarssonmalmo on July 28, 2007
I have worked with a customer with a production plant where part of their analytic requirements where about sales data and financial data. This customer had SSAS2005 cubes that worked for that and, more or less, looks like the structures in the Adventure Works cube project.
Another area with an even bigger demand for information was the manufacturing part of that company. Although there where cubes built for this area the report consumers that managed production were never given exactly what they needed.
In manufacturing you are interested in scrapping(resource waste), productivity, logistics, order stock and delivery precision. You can build SSAS cubes for that but the nature of their requirements are not working that good with predefined structures and navigational hierarchies.
They are searching for new structures and not previously known relations. What area in the MS BI platform can be of better help?
Why not try data mining?
With the sp2 release of SSAS2005 the data mining team introduced Excel2007 add ins to build data ming models directly in Excel. You can find them here. They are part of the SP2 feature pack.
You should also use this site where more information about MS data mining can be found: http://www.sqlserverdatamining.com/DMCommunity/
With the Excel2007 most of the complex side of building DM-models in BIDS has been removed. The users build the models in a dedicated SSAS2005 database that is part of the installation of the option pack.
And how do you use these Excel2007 DM add-ins? You can find an excellent virtual lab tutorial here:
It will show the well known bike buyer direct mail campaign problem and the decision tree algoritm but it is a good start to learn how the add ins work. You will follow the DM-method steps of building and training a model and, finally, use that model for prediction.
Be aware of that DM has a lot of challanges to be succesfully implemented. End users will have to be educated in how DM works. The Excel2007 add ins will not remove that problem, only bring the DM capacity to new user groups.
You will also need a data ware house with unquestionable data quality. I think DM is best suited for organizations with skillful analytic staff and a few years experience of how to run a data warehouse and an advanced BI-solution on top of that.