MLTool-BCI

Matlab/Octave Machine Learning Toolbox for linear classification with applications in Brain-Computer Interfaces (BCI). This toolbox is distributed with GPL license along a tutorial chapter Machine learning for BCI in the book.

Features

Linear models
- Linear Discriminant Analysis (LDA)
- Support Vector Machine (SVM)
- Ridge Regression (RR)
Validation strategies
- Hold-out
- Random sampling
- K-fold cross validation
- Leave-one-out bootstrap
Performance measures (classification and regression)
- Accuracy (ACC)
- Area Under the ROC curve (AUC)
- Cohen's Kappa (k)
- Means Square Error (MSE)
- Correlation coefficient (corr)
Demo Datasets
- Motor Imagery (MI) with CSP features
- P300 Speller with temporal features
- ECoG finger movement prediction dataset

Installation

Download current version here
Extract in a subfolder
Add path to Matlab/Octave with "addpath(genpath('path-to-mltool'))" to execute any function in the toolbox

Short Documentation

All functions in the toolbox contain detailled documentation with parameters definition.

classifier folder contains all the linear models estimation functions (only binary classification)
performance folder contains all the performance computation functions
validation folder contains all the pvalidation loops
figures folder contains the script used for generating figures in the chapter

Aknowlegments

Datasets

MI Dataset is extracted from dataset IIa, BCI competition IVand has been pre-precessed by Fabien Lotte.
P300 Dataset is extracted from subject APM of UAM P300 Dataset using feature extraction described in here
Finger flexion ECog dataset is extracted from dataset 4 from BCI competition IV

Code

The computation of te Area under the ROC curve is performed using svmroccurve.m that has been extracted from SVM-KM

jetW/mltool