Hello Rodrigo,
first, thanks for sharing your work!
You use oversampling on the test class in your work (pl_jaad_muster23_forecast.py) to achieve a balanced split of the classes? Is this approach valid? (In the internet it was recommended to use oversampling only in the training dataset).
In my work with the PIE and Jaad datasets, this leads to drastic improvements in the AUC, F1 and accuracy, but I am just not sure if this approach is valid.
Hello Rodrigo,
first, thanks for sharing your work!
You use oversampling on the test class in your work (pl_jaad_muster23_forecast.py) to achieve a balanced split of the classes? Is this approach valid? (In the internet it was recommended to use oversampling only in the training dataset).
In my work with the PIE and Jaad datasets, this leads to drastic improvements in the AUC, F1 and accuracy, but I am just not sure if this approach is valid.