Abstract
The brain signals recorded by EEG devices are largely developed in for biometric authentication purposes. Those signals are very informative and reliable to be classified using signal processing. In this paper, the feature extraction and feature fusion are further studied to observe their performance towards the typing tasks. The signals are pre-processed to eliminate the unwanted noise present in the signals. The feature extraction method such as Welch’s method, Burg’s method and Yule Walk’s method are applied to extract the mean, median, standard deviation and variance in the data. Nonlinear feature such as fuzzy entropy is also been extracted. The extracted features are further classified by using k-Nearest Neighbour (k-NN), Random Forest (RF) and Ensemble Bagged Tree (EBT). The performance of feature extraction and feature fusion through concatenation are recorded and compared. For comparison, the feature fusion shows a better performance accuracy rather than feature extraction. The highest percentage accuracy was produced by Burg’s method for frontal-parietal lobes feature fusion which is 95.94% using Ensemble Bagged Tree (EBT).