Hi,
In EEG datasets, I have taken 200 samples X32 channels X 300 trials. For this particular scenario, I like to do binary classification .can you please guide me on whether shall I proceed in the deep learning technique.
i) If it is in the deep learning technique, should I take feature Extraction and do classification? or without feature extraction can I do classification?
ii)In machine learning, I proceed with feature extraction, dimension reduction, and classification. If I choose deep learning without feature extraction, how can I proceed in deep learning? (i.e raw signal--> classification methods (LSTM, CNN))in between how can I introduce a new algorithm in deep learning(the stage before classification)?
Thanks,
Thenmozhi
Dear Thenmozhi Thangaraj
This paper may help to answer your question
Article Advanced Meta-Heuristics, Convolutional Neural Networks, and...
Hi,
Usually, deep learning algorithm need huge set of data, but you can overcome this challenge by using transfer learning rather than learning from scratch.
Answering your question, yes deep learning doesn't need feature extraction process because this is done automatically.
Thanks a lot Hilal sir, I am new to transfer learning. It would be better if you provide some research papers regarding this title.thanks a lot sir
You can use some data augmentation technique (Flipping, Rotation, Translation, etc ...) to increase your number of sample features.
First of all, increase the number of dataset by using state of art data augmentation technique Like- cutmix or others.
I. Can try with transfer learning for feature extraction then go with classification & actually deep learning used with the extracted features. So there is no point of view without it.
ii. Try to change the number of layers in your cnn/lstm model & check their performances . Then you can find the best one
hi, look at the NN architecture named as 'Siamese neural network', maybe it will useful for you
- Badges
- Science topic