multiclass dataset

Question

multiclass dataset

Closed this issue 7 years ago · 7 comments

your proposed hybrid approach is it applicable for multiclass dataset?

Answer 1 · 2018-01-05T19:58:20.000Z

Oversampling more than two classes is not currently possible.

Answer 2 · 2018-01-05T20:42:41.000Z

Thanks for your quick answer i am working on a multiclass imbalanced dataset and i am trying to balance it wit SCUT https://www.researchgate.net/publication/301453161_SCUT_Multi-Class_Imbalanced_Data_Classification_using_SMOTE_and_Cluster-based_Undersampling
But i've faced some problem in implementation because no further details about the strategy implemented .

Answer 3 · 2018-01-05T20:44:12.000Z

If you're ready we can collaborate together and propose a new hybrid technique to overcome multiclass data

Answer 4 · 2018-01-08T10:51:10.000Z

Thank you, but I'm not interested at the moment. In the future, multi-class support may be added to k-means SMOTE.

Answer 5 · 2018-02-06T17:11:20.000Z

Multi-class support is now available with release 0.1.0.

Answer 6 · 2018-02-06T21:36:12.000Z

Thanks Felix where can i find it and do you have any description of your extension

Answer 7 · 2018-02-09T17:20:36.000Z

@rawiasammout, you can simply pip install kmeans-smote as shown in the readme. The latest version (0.1.0) implements the multi-class case.

To use kmeans-smote on multi-class data, there is nothing special to consider, just call fit_sample on your multi-class data and tune the algorithm's hyperparameters (see docs) to your needs. There is also a small test implemented to oversample a multi-class toy dataset.

From an algorithmic perspective, the second step of the algorithm (see readme > about) is repeated for each minority class. The paper referenced in the readme should provide you with a thorough explanation of how the algorithm works for the binary case - the multi-class case is only a small modification.