3
$\begingroup$

In Excel's data mining tools there is a "Key Influencers" tool which will look at a dataset which is perhaps customers and whether or not they converted to a given goal (e.g. a flag that equals 1). It then tells you the most influential factors in reaching that goal,( e.g. Gender=Male and Age=30-45). What would be equivalent algorithm in R to achieve a similar outcome.

$\endgroup$
2
  • 1
    $\begingroup$ What approach does the excel command use? That might be useful to know. $\endgroup$
    – charles
    Commented Aug 20, 2014 at 15:53
  • 1
    $\begingroup$ My guess is it's really something pretty ad hoc, like finding those variables having a correlation with the outcome variable greater than some threshold such as 0.5. It seems like trying to duplicate something Excel does in R is kind of an empty cause: it's so easy to do something much better. $\endgroup$
    – Russ Lenth
    Commented Aug 20, 2014 at 17:44

1 Answer 1

3
$\begingroup$

The algorithm used in the "Analyse Key Influencers" tool in excel is a variation on the Naive Bayesian algorithm for classification. The explanation from Microsoft may be found here:

http://msdn.microsoft.com/en-us/library/dn282341.aspx

The naive Bayes algorithm has been implemented in several packages, though most notably the e1071 package by David Meyer. It can be found on CRAN and a decent tutorial on implementing the algorithm may be found here:

http://en.wikibooks.org/wiki/Data_Mining_Algorithms_In_R/Classification/Na%C3%AFve_Bayes

I would caution you to ensure that your predictor variables are independent and to avoid multicollinearity in your models, as the algorithm ignores dependencies and is thus "naive" or "stupid".

$\endgroup$
2
  • 2
    $\begingroup$ If Naive Bayes was an option then logistic regression is a better one, with weaker requirements and no pesky conditional independence assumption. $\endgroup$ Commented Aug 20, 2014 at 18:42
  • $\begingroup$ Multicollinearity is still a problem with logistic regression! $\endgroup$ Commented Aug 20, 2014 at 19:31

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.