Investigation on EEG Features and Classification Methods for Brain Computer Interface

Currently, the number of disabled people is increasing rapidly, requiring the development of applications with the aim to establish a communication between the human brain and external assistance devices through a computer, also known as brain computer interface (BCI). Most commonly, EEG signals are used for BCI system because they have high temporal resolution and are noninvasive. In this study, we focused on classifying EEG signals acquired through two experiments, (1) using two channels C3, C4 and (2) by six channels attached on the motor cortex to distinguish the left and right hand movement imaginations. For the classification task, k-Nearest Neighbor (kNN), Artificial Neural Network (ANN), Support Vector Machine (SVM), and Decision Tree were investigated. Furthermore, we combined the Principal Components Analysis (PCA) method and the Common Spatial Patterns (CSP) method to reduce redundant features in the recorded signals from 6-channel experiments in order to enhance the classification accuracies. The overall accuracy of recognizing two hand movement imaginations was 85% for KNN, 97% for ANN, 97% for SVM and 100% for Randomized Forest respectively with ratio 50% train data and 50% test data. These results show the potential of further developing applications in BCI electric wheel chair. The data is recorded by self-implement EEG system using ADS1299.