EEG signals classification using SVM and LDA

Fall '20

technical skills: custom Linux distro development, EFI, Lua, C
soft skills: cross-functional interaction with internal teams and external vendors

This project is to classify between 2 actions "eating" and "writing" using EEG signals.The participant was instructed to perform two types of activities: writing name and bringing their hand to mouth (as normally done when eating). The EEG data was recorded with a sampling frequency of 500Hz and stored in the BioRadioData variable (8 channels). The participant was wearing a 3-axes accelerometer (IMU) recording acceleration of hand motion during each activity

The goal for this activity is to use the IMU data to label the EEG signals due to the fact that EEG signals are harder to track down. Initially, I filtered out the EEG signals using low pass filter. Then I splitted the data to 75%-25% between train and test set. After that, I used Common Spatial pattern for extracting the features before putting the data to train our machine learning model.

For our classification models, I tried LDA and SVM with 2nd and 6th order for better comparison. As a result, LDA is a better choice with higher accuracy in the test set than SVM.