Affiliations: [a] School of Mechatronic Engineering, Universiti Malaysia Perlis, Perlis, Malaysia | [b] Malaysian Spanish Institute, Universiti Kuala Lumpur, Kulim Hi-TechPark, 09000 Kulim, Kedah, Malaysia
Correspondence:
[*]
Corresponding author: Sathees Kumar Nataraj, School of Mechatronic Engineering, Univerisiti Malaysia Perlis, Kangar, Malaysia. E-mail: satheesjuly4@gmail.com.
Abstract: In this research work, a simple Electroencephalogram (EEG) based imagery vocabulary classification system has been developed for the Differentially Enabled (DE) communities, to communicate their needs with the outside world. The proposed communication system consists of a simple data acquisition protocol, which includes the basic needs of DE patients in their daily life, such as Food, Water, Toilet, Help, Aircon, Tv and Relax. The EEG signals for each task are recorded from ten subjects using a standard wireless EEG amplifier from eight different electrode positions. The recorded brain wave patterns are pre-processed and segmented into four frequency bands, namely Delta (δ), Theta (θ), Alpha (α) and Beta (β). A simple feature extraction technique using cross-correlation (r) estimation has been proposed to extract the coefficients between any two frequency bands. Similarly, six permutation sets of four frequency bands for each electrode position are framed and the statistical features such as minimum (min), maximum (max), mean (μ), standard deviation (σ), skewness (G) and kurtosis (K) are computed to form the feature sets. The extracted feature sets are classified using three different supervised non-parametric classification methods, namely, k-Nearest Neighbor (k-NN), Multilayer Neural Network (MLNN) and Probabilistic Neural Network (PNN). Further, the classification models are compared and from the results it is observed that the k-NN classifier hits the highest classification accuracy of 90.24% using max feature set.
