I'm looking for imbalanced classification datasets to experiment with using synthetic data, ideally with a minor class of less than 10%. Does anyone know specific ones? Should be open to the public with no limitation of use.
Add a comment
|
2 Answers
Most multi-class datasets can be turned into skewed binary classification datasets.
For example, the default scikit-learn digits dataset contains ~10% 1 and ~90% not-1 images.
That being said, the UCI Machine Learning Repository hosts many datasets that are skewed, one that's quite skewed is this SMS spam dataset: https://archive.ics.uci.edu/ml/datasets/SMS+Spam+Collection
Another organisation that hosts free but not open datasets is kaggle, one of their datasets that's very skewed is the following:
https://www.kaggle.com/c/jigsaw-toxic-comment-classification-challenge
-
1its not open data if you have to register, agree to terms, and give your phone number for verification.– albertJul 12, 2018 at 16:19
-
This specific dataset should be open as it consists of Wikipedia comments, however, I'll verify and possibly correct when I get home.– SwierJul 12, 2018 at 16:46
-
-
1
Also found this one: https://www.kaggle.com/mlg-ulb/creditcardfraud - also taken from Kaggle.