I posted this question on another place, but I want to get many tips, so I post here too.

I am building deep learning classification model in bioinformatics. I made training dataset by merging 12 datasets, each dataset has some common features, and some common label, and there are uncommon features and labels. So I outer-joined the data for training.

But this training data is quite imbalanced itself (31%, 20%, 15%, 7%, 6%, 1%, ... 0.1 % for each class), this is quite challenging to me already, but even more, I can't even suppose test dataset's distribution (which means test dataset or real-world data that I should predict can be very different from training dataset distribution of class,

as far as I know deep learning classification model trained by specific distribution can predict well if only data to predict has same distribution.),

test dataset now I am using is quite imbalanced, has different distribution compared to training dataset, even more some of labels are missing.

What kind of approach, and deep learning model should I take? Or is there any good paper for these kind of problems?



Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.