1
$\begingroup$

I'm about to work with concept drift problem in data streams.

I need to start with real data sets from UCI machine learning repository and add to them concept drift (in attributes domain).

Do you know tools that can help me with this? Maybe someone knows it is possible to use MOA[1] framework for this?

What is best way to approach this if I have to do it by myself?

[1] http://moa.cms.waikato.ac.nz/

$\endgroup$
1
$\begingroup$

It sounds like you are trying to generate a synthetic data set, by artificially adding fake drift.

This is a bad idea. Don't do this.

If you manage to find some algorithm that works well on your synthetic data set, then all you will know is that your algorithm works on a fake data set -- but you'll have no idea whether it is useful in practice.

Instead, I suggest that you try to find some real-world data sets that contain concept drift, and use them to work with. If you're not sure how to do that, you might not be ready to do research in this area yet: you might want to spend some time talking to your advisor/mentor, or spend a bunch of quality time in the library reading papers in the literature that talk about the concept drift problem.

$\endgroup$
  • $\begingroup$ First of all - I can't find data sets with drift for regression task. Internet is full od data sets with drift for classification tasks. Synthetic data is first step and in this kind of data I know how it should behave and did concept drift detection algoritm worked or not. $\endgroup$ – Mr Jedi Mar 23 '15 at 8:16

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.