2
$\begingroup$

I have a dataset as (var1, var2, out), where the ordered pair <var1, var2> gives out. Most of the frequent pattern mining algorithms like the Apriori and FP growth algorithms does not preserve the order of var1 and var2.

Which are some of the available pattern mining algorithms (may also be a NN trick), to find association rules between ordered pair <var1, var2> and output variable out?

Thanks.

$\endgroup$
  • $\begingroup$ What about KNN? $\endgroup$ – Mikhail Berlinkov Nov 13 '18 at 18:28
  • $\begingroup$ Can you please describe more on the suggested approach? As it is a big data problem, not sure what would be a reasonable value for k in KNN. $\endgroup$ – user3243499 Nov 13 '18 at 18:32
  • $\begingroup$ Maybe KNN is not the best choice if you deal with Big Data. What is the distributions of your features and outcomes? $\endgroup$ – Mikhail Berlinkov Nov 13 '18 at 21:57
  • $\begingroup$ in general pattern analysis is done with Markov models and also based on data, industry also algorithm changes $\endgroup$ – sai saran Nov 14 '18 at 14:46
  • $\begingroup$ Can you give us more information? You have only two input variables? Or the variables are sets? The outcome is discrete? $\endgroup$ – rapaio Nov 15 '18 at 6:09
0
$\begingroup$

Assuming you only have these two features (var1, var2), you might want to: * Create one-hot encoded features for each variable under each position. * Add a column on which variable is first (e.g. two columns - likely to work with trees but not with anything else). * Take each possible combination of variables and use that as your only input (e.g. you then take the average of out for that combination, perhaps adding some prior or smoothing).

As the comments mentioned, if out is some discrete event, maybe you'd want to instead look at Markov models.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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