In this Project, I extracted some features from BCI2003 data and selected the best features for classifying the signals using PSO (particle swarm optimization). This project was done for the EE-SUT computational intelligence course.
First, I maximized the variance between the classes using CSP filters. Then, I extracted the following features from the trials. Finally, I found the best features which have the maximum classification accuracy using PSO and an MLP network.
- Mean frequency
- Median frequency
- Total power of the channels
- Power of delta, theta, alphas, beta, and gamma frequency bands
- Entropy
- Lyapunov exponent
- Average of differentiate of trials
- Skewness
- Kurtosis
- Phase of the FFT coefficients
Final accuracy with 5-fold validation was 87%.
This project uses Matlab for feature selection and extraction and uses python for MLP classification. You can install the required libraries for python using pip install -r requirements.txt. Also, first, change the path to pyenv in the first lines of main.m.