2

There are many oversampling methods for categorical labels (for example SMOTE and Rose, etc.).

But, are there oversampling method for numerical labels (the thing that I want to predict with my features), in the sense that it applies something similar SMOTE or Rose for the label that is highly skewed? (I don't want to normalize or scale my label/class) for regression types of tasks?

jennifer ruurs
  • 133
  • 5
  • Have you gone through the linear regression diagnostics in the normal way? http://www.sthda.com/english/articles/39-regression-model-diagnostics/161-linear-regression-assumptions-and-diagnostics-in-r-essentials/. You can transform the data if necessary e.g. log etc. – Christopher John Nov 02 '19 at 06:53

1 Answers1

1

There is some research on this topic. However, it remains less explored than its classification counterpart, as you have likely encountered.

I might suggest the paper cited below (and the conference it was presented at http://proceedings.mlr.press/v74/) depending on how interested you are in understanding it from a research perspective. I really appreciated the introduction of Gaussian noise in generating the synthetic observations.

If you're more interested in a practical solution, the first author has an R implementation available on her Github page. https://github.com/paobranco/SMOGN-LIDTA17

Also, I am currently developing an entirely Pythonic implementation of the SMOGN algorithm that will be available shortly. https://github.com/nickkunz/smogn

If you need a fast and intuitive solution to a highly skewed distribution, a common method is just using the log of your variable. Although, I understand that this has its obvious limitations. I hope this helped.

Branco, P., Torgo, L., Ribeiro, R. (2017). "SMOGN: A Pre-Processing Approach for Imbalanced Regression". Proceedings of Machine Learning Research, 74:36-50. http://proceedings.mlr.press/v74/branco17a/branco17a.pdf.

Nick Kunz
  • 86
  • 1
  • 3