Creating an Imbalanced Dataset

Question

I would like to have my trained model tested on an imbalanced dataset. Is there any algorithms available to generate synthetic data from a balanced labelled dataset (spam/non-spam)?

You can always unbalance any data set by simply undersampling one class. — user2974951, Sep 19 '18 at 10:54

score 8 · Answer 1 · answered Sep 19 '18 at 11:42

8

Try SMOTE, its an algorithm used for over-sampling. It creates synthetic samples from the class you want over-sampled.

You can use this to create any number of samples you need.

answered Sep 19 '18 at 11:42

Mary93

403
2
9

1

can SMOTE be used for under-sampling as well? – Stuart Peterson Sep 19 '18 at 13:12
Well, you can obtain undersampling of class A by oversampling class notA ... – kjetil b halvorsen Sep 19 '18 at 14:30
3

@StuartPeterson No, SMOTE is an over-sampling algorithm, but there are [many other](https://imbalanced-learn.readthedocs.io/en/latest/api.html#module-imblearn.under_sampling) under-sampling algorithms – Mary93 Sep 24 '18 at 18:53

Creating an Imbalanced Dataset

1 Answers1