Feature extraction techniques for cognitive stimuli-based electroencephalogram signals: an experimental analysis

Abstract

Over the past decades, brain-computer interface (BCI) has gained a lot of attention in various fields ranging from medicine to entertainment, and electroencephalogram (EEG) signals are widely used in BCI. Brain-computer interface made human-computer interaction possible by using information acquired from EEG signals of the person. The raw EEG signals need to be processed to obtain valuable information which could be used for communication purposes. The objective of this paper is to identify the best combination of features that could discriminate cognitive stimuli-based tasks. EEG signals are recorded while the subjects are performing some arithmetical based mental tasks. Statistical, power, entropy, and fractional dimension (FD) features are extracted from the EEG signals. Various combinations of these features are analyzed and validated using random forest classifier, K-nearest neighbors (KNN), multilayer perceptron, linear discriminant analysis, and support vector machine. The combination of entropy-FD features gives the highest accuracy of 90.47% with the KNN algorithm when compared to individual entropy and FD features which achieves 79.36% with random forest classifier, multilayer perceptron, and 82.53% with linear discriminant analysis, respectively. Our results show that the hybrid of entropy-FD features with KNN classifier can efficiently classify the cognition-based stimuli.