T-Test

sudheendra
sudheendra New Altair Community Member
edited November 5 in Community Q&A
Hi All,

How to start exploring with T-Test and Anova? I would like to use performance vector for my sample dataset.which one will be better to start with.

Regards
Sekhar
Tagged:

Answers

  • land
    land New Altair Community Member
    Hi,
    you have to provide two performance vectors and then apply the T-Test or Anova operator on it. It will give you the probability, that they are not sigificantly different.

    Greetings,
      Sebastian
  • crappy_viking
    crappy_viking New Altair Community Member
    Hi,

    About Tests, would you have a look here ? http://www.statsnetbase.com/ejournals/books/book_summary/summary.asp?id=1658
    The "henry" normality test seems interesting because it can help in deciding whether to use Neural models or linear models. If you know other normality tests, let me know...

    C.V.
  • AnneG
    AnneG New Altair Community Member
    Hello there,
    I guess this question fits in here ... For quite a long time I have been trying to figure out how to perform t-tests on my data. I hope you will give me a hint on this. I have three classes and I want to compare (in pairwise manner) each attribute within one class with another class. So, is there a significant difference between class 1 and 2 concerning attribute A and so on. Thus, this is actually no comparison of performances, but I know that I need performance vectors as input for the t-test operator. However, I cannot think of a way of getting to this performance vector from my example set.

    Greetings,
    Anne
  • land
    land New Altair Community Member
    Hi,
    we have the AnovaMatrix doing something similar. It calculates the significance of difference between the values of all numerical attributes, based upon a grouping defined by all nominal attributes.

    Might this help?

    Greetings,
      Sebastian
  • AnneG
    AnneG New Altair Community Member
    Hello,

    yes ANOVA is alright, however, as soon as there are more than two classes to compare, ANOVA does not tell you WHERE the differences lie. Say, we have attribute X where class 1 and 2 are different, but there is no difference between 1 and 3 and between 2 and 3. That is why one would need post hoc tests such as Scheffé or Bonferroni.
    But I see, the t-test and anova operators in RapidMiner are not meant to be used like this. In any case your tool is still great ;-)

    Kind Regards,
    Anne
  • land
    land New Altair Community Member
    Hi,
    feel free to implement such a operator. I will be happy to put it into the core and provide it's functionality to all users. Shouldn't be to much of a problem, if you know the algorithm.

    Greetings,
      Sebastian

Categories