EEG signals classifications of motor imagery using adaptive neuro-fuzzy inference system and interval type-2 fuzzy system

Abstract

Brain computer interface (BCI) techniques are used to help disabled people to translate brain signals to control commands imitating specific human thinking based on electroencephalography (EEG) signal processing. The paper tries to accurately classify motor imagery imagination tasks, e.g., left and right hand movement. The paper utilises different methods for such classification including: (1) adaptive neuro fuzzy inference system (ANFIS); (2) K-nearest neighbour (KNN); (3) linear discriminant analysis (LDA) and (4) interval Type-2 fuzzy system (IT2-FS) classifiers. In addition, with ANFIS approach, different clustering methods are examined such as Subtractive clustering, fuzzy C-means clustering and K-means clustering. At the same time, subtractiveType-2 clustering is applied to the received signals. The paper focuses on three different features which are AR coefficients, Band Power Frequency, and common spatial pattern (CSP). The classification accuracies with two optimal channels C3 and C4 are investigated.