BCI-emotion-recognition

A fundamental exploration about EEG-BCI emotion recognition using the SEED dataset & dataset from kaggle.

Dataset

• Emotion Recognition Dataset on Kaggle: https://www.kaggle.com/datasets/birdy654/eeg-brainwave-dataset-feeling-emotions

• SEED: https://bcmi.sjtu.edu.cn/home/seed/seed.html

Models

• GRU
• LSTM
• Transformer: EEG-conformer: https://github.com/eeyhsong/EEG-Conformer

Results

Warning

Thanks for AllGGI to point out the bug in my original impl.. I fixed the bug and re-do the experiments(only the subject-dependent ones). Sadly, the results were pretty bad. Therefore, I suppose this work isn't a successful practice, and hidden bugs may remain in the codings.

note:

• Models are trained & tested on SEED dataset

● subject-dependent: train & test on subject1, 100 epochs

● subject-independent: train & test on a mixed dataset of all subjects, 50 epochs

File Descriptions

.
├── base.py # the base helper functions
├── conformer.ipynb # conformer on SEED 
├── conformer-sub1.ipynb # conformer on SEED, subject1
├── eegconformer.py # the implementation of conformer
├── emotions.csv # the Kaggle dataset
├── gru.ipynb # gru on Kaggle dataset
├── gru-seed.ipynb # GRU on SEED 
├── gru-sub1.ipynb # GRU on SEED, subject1
├── LSTM-seed.ipynb # LSTM on SEED
├── LSTM-sub1.ipynb # LSTM on SEED, subject1
├── model_gru # best model state dict 
├── model_LSTM # best model state dict
├── model_transformer # best model state dict
├── Preprocessed_EEG # the SEED dataset 
└── README.md