Sign in
 
Home20122008 R2200820052000LibraryForums
SQL Server >
> 
> 
best methods/techniques for data associations

Pregunta best methods/techniques for data associations

  • Wednesday, February 20, 2013 8:22 PM
     
     
    Sign In to Vote
    0

     Hi all...

    I'm looking for an data mining approach that could be used to help associate data based on weighted metadata of particular data elements but with the added feature that one could 'pivot' on an element to observe additional associations.  I'm mixing methods and function here, so let me try to paint what I am after with an example.  

    Let's say a loan officer plugs in 15 data elements to a form for a bank.  I could imagine that if the bank program used a hybrid data classification technique (back propagation on historical data showing parameters and output of poor loan performance) new data could be evaluated as pass of fail.  But I'd like to be able to weigh in on what value caused the failure, pivot on it (select it) and look at all the associations with that bad factor...perhaps there is a relationship with other data elements that tell us something.  Maybe if it was the person's poor credit score that was very low, I could find out that everyone in the system with that same credit score also happened to have the middle name 'Larry'.  Okay, bad example, but hope you see what I am after.  Point is I'm looking for a way to mark data with certain weights as bad or good, then be able to pivot on that element to churn new information about that particular element.  Sort of combining different features of SQL and Excel???  


     

All Replies

  • Thursday, February 21, 2013 8:58 AM
     
     
    Sign In to Vote
    0

    So you want to find simarilaties with people who have a good / bad score and associate them into the same group?

    If this is what you are asking then you dont need to worry about this as the model will do this for you, you would need to ensure you feed the variables into the model (in your example, person's name and their score) and the model will calculate this for you.

    Is this what you were asking?

     
  • Friday, February 22, 2013 1:14 PM
     
     
    Sign In to Vote
    0

    Well, not really - I'm trying to identify a technique to use for associating data elements that would allow me to pivot on one parameter in a record and then find all records that have association with that particular parameter. 

    The loan example turns out to be a bad analogy.  Perhaps another analogy would be crime records.  Say an incident is recorded and one of the parameters of that record turns out to be an address.  I could select the address, then find out that address has been involved in many other incidents. I could select the owner of the property, then show all associations to that property owner (perhaps agents entered in intel previously on those associations).  I'm just trying to determine which type of mining model, data association model, might support this type of logic.