clustering and decision tree

Discussion in 'microsoft.public.sqlserver.datamining' started by z_abderrahman, Oct 27, 2004.

  1. Hi;
    I am undertaking an CRM analysis for a company.
    I want to know :
    1- (in depth) the algorithms used in Clustering and Decision tree.
    2- How can I select the "good cluster" and the best DT.
    3- How to interpret the results.

    Note: I have read the tutor.

    I appreciate your help.

    Yours Sincerely.

    **********************************************************************
    Sent via Fuzzy Software @ http://www.fuzzysoftware.com/
    Comprehensive, categorised, searchable collection of links to ASP & ASP.NET resources...
     
    z_abderrahman, Oct 27, 2004
    #1
    1. Advertisements

  2. See below

    --
    Peter Kim
    This posting is provided "AS IS" with no warranties, and confers no rights.

    There is an FAQ for this. Check out
    http://www.sqlserverdatamining.com/DMCommunity/FAQ/default.aspx - look for
    item 16).
    Since clustering is unsupervised task, it's very subjective to know which
    model is the best. In general, a good cluster model should be high level of
    coherency inside each cluster and high level of dissimilarity amoung the
    clusters. I believe there are quite a few papers to address this issue. DT
    is a supervised algorithm, meaning that you have the test data that have
    answers. Runing a prediction DMX over the test data can measure the accuracy
    of all candidate models.

    SQL 2000 data mining doesn't have lots of UI tool to help this task. The
    cluster viewer shows the distribution of each cluster from which the user
    can "intuitively" get insights on it. The tree viewer shows the tree
    content. However, SQL 2005 DM (beta2) has added quite a few UI support for
    this. The cluster viewer show profile of each cluster, dependency among the
    clusters and cluster discrimination. Also, SQL 2005 DM has added Lift Chart
    from which you can easily see which candicate model yields best lift for
    test data set.
    Could you please narrow down your question?
     
    Peter Kim [MS], Oct 27, 2004
    #2
    1. Advertisements

Ask a Question

Want to reply to this thread or ask your own question?

You'll need to choose a username for the site, which only take a couple of moments (here). After that, you can post your question and our members will help you out.