Time Series

Discussion in 'microsoft.public.sqlserver.datamining' started by pras kumar, May 23, 2005.

  1. pras kumar

    pras kumar Guest

    I am just curious to know about time series.Where can i find a good
    source to learn about that.

    pras kumar, May 23, 2005
    1. Advertisements

  2. pras kumar

    MC Guest

    1. Advertisements

  3. pras kumar

    pras kumar Guest

    Thanks that information was useful.I have some questions

    1)Why is feature selection needed?If we give more attributes to the
    mining model..It is like giving more information to create the mining
    model..Then why is it neccesary to do feature selection?

    2)I used microsoft decision tree in sql server 2000 to create the mining
    model for my project.From my data mining project i learnt that in order
    to get a accurate tree we have to create various trees and then decide
    which tree is most accurate........

    does it mean that the mining model that we create using the neural
    network approach is always accurate??Because i studied that in neural
    network apprach backpropogation approach is used .so the error that is
    obtained is very minimal..
    pras kumar, May 24, 2005
  4. 1: Feature selection is necessary if you want the problem to finish in any
    reasonable time. Using our Time Series algorithm, the features are T0
    (output), T-1, T-2, T-3,T-4,T-5,T-6,T-7,T-8. If you add any periodicity
    hints (which the algorithm finds automatically by default), you get features
    T-hint through T-hint*8, eg. T-12 through t-96. Since our Time Series
    allows for multiple series, you have the same features for series R, S, Q,
    U, W, X, Y and Z, etc. To summarize you get many more features than it
    makes sense to consider. Feature selection attempts to reduce the search
    space without reducing model accuracy.

    2: No, neural nets are not necessarily more accurate than trees. This is
    why we allow the comparison of models in our accuracy charts in SQL 2005 so
    you can easily see which performs best.
    Jamie MacLennan \(MS\), May 24, 2005
    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.