ABSTRACT
Electroencephalography (EEG)-based brain-computer interface systems are designed to serve as a communication channel to the people with disabilities to interact with their external environment. Decoding imagined speech from one’s mind has become the most recent field of interest as it consists of the pronunciation of words internally. However, it is difficult to decode the imagined words due to the complex nature of the EEG signals. The objective of the study is to classify the imagined words and phonological categories and evaluate the feasibility of this task based on the accuracy arising from them. In this regard, EEG data are preprocessed to minimize the effect of artefacts and noise. EEG signal is further decomposed by wavelet transform and empirical mode decomposition to extract meaningful information and evaluate statistical features. The Kruskal-Wallis test is performed to select the highly discriminative features. Different classification algorithms, k-nearest neighbors, decision tree, support vector machine, and ensemble bagged tree, are employed for the classification. The proposed method has obtained the maximum binary classification accuracy of 94.69% for phonological categories and 53.07% for multiclass classification of imagined words. The results obtained by the proposed method overcome the baseline results on the same database.
