Authors: Yundong Wang, Zimu Li, Haoran Zhang, Yadi Deng
This project applied machine learning techniques to the P300 speller classification challenge at Kaggle BCI Challenge @ NER 2015. The P300 speller is a brain-computer interface paradigm that allows one to input text or commands to a computer via brain activity measured by electroencephalography (EEG).
xDAWN “Covariance algorithm and Tangent Space algorithm” which uses the spatial correlations of the simultaneously recorded EEG as features for the error detection classification are implemented.
We applied multiple machine learning techniques (SVM, Logistic Regression, Random Forest, etc) and deep learning model dedicated for EEG data (EEGNET). Lastly, we built a StackNet model that achieved the 4th best AUC (0.803) result compared to the candidates pool.
The presentation slides can be found here.
Each session include 60 target stimulus, however, the last session of each subject contains 100 target stimulus. Which makes 340 target stimulus for each subject
Data is collected at 200 Hz across 26 subjects (16 for training, 10 for testing). Each subject participated in 5 different sessions.
0 or 1 for bad or good feedback, respectively. Bad feedback is when the selected item is different from the expected item. Good feedback is when the selected item is similar to the expected item.
The Kaggle description and dataset can be downloaded here.
To run python code, please download data from https://www.kaggle.com/c/inria-bci-challenge/data and put train and test directories under data. Python version: 3.6
To run preprocess.py, install pyriemann packages using pip.
To run EEGNET.py, install EEGModels at https://github.com/vlawhern/arl-eegmodels and required TensorFlow packages.
To run StackNet.py, install pystacknet at https://github.com/h2oai/pystacknet, install LightGBM and XGBoost using conda or pip.
|
|
|
|
Attached below are the results of models we applied to the pre-processed EEG data.
-
Rivet, B.; Souloumiac, A.; Attina, V.; Gibert, G., "xDAWN Algorithm to Enhance Evoked Potentials: Application to Brain–Computer Interface," IEEE Transactions on Biomedical Engineering, vol.56, no.8, pp.2035,2043, Aug. 2009
-
A. Barachant, S. Bonnet, M. Congedo and C. Jutten, “Multiclass Brain-Computer Interface Classification by Riemannian Geometry,” in IEEE Transactions on Biomedical Engineering, vol. 59, no. 4, p. 920-928, 2012 PDF
-
A. Barachant, S. Bonnet, M. Congedo and C. Jutten, “Classification of covariance matrices using a Riemannian-based kernel for BCI applications“, in NeuroComputing, vol. 112, p. 172-178, 2013 PDF
-
Lawhern, Vernon J. EEGNet: A Compact Convolutional Network for EEG-based Brain-Computer Interfaces, 23 Nov. 2016, arxiv.org/abs/1611.08024v4. Accessed 16 Mar. 2019 PDF