Sign in
 
Home20122008 R2200820052000LibraryForums
SQL Server >
> 
> 
Cross-validation for Microsoft Association Rules 2012, data mining accuracy comparison

问题 Cross-validation for Microsoft Association Rules 2012, data mining accuracy comparison

  • Thursday, March 21, 2013 12:25 AM
     
     
    Sign In to Vote
    0

    Hi all,

    I am trying to create a meaningful cross-validation for Microsoft Association Rules. I have the classic shopping basket analysis and would like to use cross validation to measure and compare the accuracy of 4 mining models in predicting association rules.

    The mining models in question use Decision Tree, Naive Bayes, Clustering and Association algorithms respectively. When I use a single target attribute as an input parameter, I get 4 identical results for the models... I don't quite understand what is going wrong.

    Thanks for sharing any knowledge.

     

All Replies

  • Friday, March 22, 2013 8:33 AM
    Moderator
     
     
    Sign In to Vote
    1

    Hi sh1361m,

    If the target attribute is a nested column, or a column in a nested table, you must type the name of the nested column using the format <Nested Table Name>(key).<Nested Column>. If the only column used from the nested table is the key column, you can use <Nested Table Name>(key). After you select the predictable attribute, Analysis Services automatically tests all models that use the same predictable attribute. If the target attribute contains discrete values, after you have selected the predictable column, you can optionally type a target state, if there is a specific value that you want to predict. Please refer to: http://msdn.microsoft.com/en-us/library/bb895174.aspx
    http://msdn.microsoft.com/en-us/library/ms167032.aspx

    Thanks,
    Eileen

    If you have any feedback on our support, please click here

    Eileen Zhao
    TechNet Community Support

     
  • Friday, March 22, 2013 5:06 PM
     
     
    Sign In to Vote
    0

    Hi Eileen,

    Many thanks for your response. I tried your suggestion by using the format <Nested Table Name>(key) as my nested column only has the key column (which is also selected as input and predict). My Target Attribute is therefore : SFDM_OrderLines(line_key) I am getting below error message:

    Error (Data mining): The specified mining structure does not contain a valid model for the current task.

    Alternatively, I also tried copying and pasting one of the actual attribute names from the node list. SFDM Order Lines(WAV) this shows the table name and one of the products 'WAV"

    Doing so, the process runs and returns some results for both Association and Decision Tree algorithms. I am trying to test how well each of these models work for making association between various products. Is my 'target value' valid in this case? Does it mean that if I use (SFDM Order Lines(WAV)) as the target value, the cross validation is only checking for nodes containing this single product? What about the association between more than 1 product? Is there any way of testing for "true" association algorithm?

    many thanks

    see below snapshots of some parts of the mining model