A Naïve Bayesian approach to lower limb classification from EEG signals

Lower limb movement decoding is the primary objective for designing Brain-computer interface (BCI) controlled leg prosthesis. In this paper the main objective is to classify left-right leg movement directly from electroencephalography (EEG) signals by probabilistic Naive Bayesian approach. Wavelet based decomposition has been taken as feature extractor to take care of non-stationary nature of brain waves, that have been recorded from 12 healthy subjects. The proposed method achieved average accuracy of 78.33% with 11ms of execution time. Specificity, sensitivity, type 1 and type 2 error rate have also been determined for each case. Results of the classifier is compared with other standard classifiers and statistically validated by Friedman Test. Novelty of the paper lies in the fact that it considers the both frequency and spatial domain (location in time) features of EEG signal without sacrificing accuracy and very low execution time makes it feasible for real time application. It also shows the Naïve Bayes classifier with uniform prior probability is better classifier than standard Naïve Bayes classifier in recognizing left-right lower limb movement.