Hi Rajesh Neupane

The undersampling technique must be applied on the training set only. You need to make sure that none of the instances in the training set is located in the test set data during the learning process.

Regards,

Dr. Samer Sarsam

Undersampling can be done before or after the train-test split, as long as it is only applied to the training data. If you undersample before the train-test split, you will have a smaller training dataset but you will not introduce any data leakage. If you undersample after the train-test split, you will have a larger training dataset but you will need to be careful to not use the test data to inform your undersampling decisions.

Both oversampling and undersampling can be useful techniques for addressing imbalanced datasets, but it is important to carefully consider the potential effects on the training data and the potential for introducing data leakage. It may also be helpful to try both oversampling and undersampling and compare the performance of your model using each approach.

Dear Samer Sarsam and Abderrahmane Boudribila thank you for your insights, it was really helpful to complete my masters thesis.

with regards,

Rajesh Neupane