The classification of motor imagery response: an accuracy enhancement through the ensemble of random subspace k-NN

The classification of motor imagery response: an accuracy enhancement through the ensemble of random subspace k-NN Mamunur Rashid1, Bifta Sama Bari1, Md Jahid Hasan2, Mohd Azraai Mohd Razman2, Rabiu Muazu Musa3, Ahmad Fakhri Ab Nasir2,4 and Anwar P.P. Abdul Majeed2,4 1 Faculty of Electrical and Electronics Engineering Technology, Universiti Malaysia Pahang, Pekan, Pahang, Malaysia 2 Innovative Manufacturing, Mechatronics and Sports Laboratory, Faculty of Manufacturing and Mechatronic Engineering Technology, Universiti Malaysia Pahang, Pekan, Pahang, Malaysia 3 Centre for Fundamental and Continuing Education, Universiti Malaysia Terengganu, Kuala Nerus, Terengganu, Malaysia 4 Centre for Software Development & Integrated Computing, Universiti Malaysia Pahang, Pekan, Pahang, Malaysia ABSTRACT Submitted 21 October 2020 Accepted 6 January 2021 Published 2 March 2021 Corresponding authors Mamunur Rashid, Anwar P.P. Abdul Majeed, Academic editor Robertas Damaševičius Additional Information and Declarations can be found on page 25 DOI 10.7717/peerj-cs.374 Copyright 2021 Rashid et al. Distributed under Creative Commons CC-BY 4.0 Brain-computer interface (BCI) is a viable alternative communication strategy for patients of neurological disorders as it facilitates the translation of human intent into device commands. The performance of BCIs primarily depends on the efficacy of the feature extraction and feature selection techniques, as well as the classification algorithms employed. More often than not, high dimensional feature set contains redundant features that may degrade a given classifier’s performance. In the present investigation, an ensemble learning-based classification algorithm, namely random subspace k-nearest neighbour (k-NN) has been proposed to classify the motor imagery (MI) data. The common spatial pattern (CSP) has been applied to extract the features from the MI response, and the effectiveness of random forest (RF)-based feature selection algorithm has also been investigated. In order to evaluate the efficacy of the proposed method, an experimental study has been implemented using four publicly available MI dataset (BCI Competition III dataset 1 (data-1), dataset IIIA (data-2), dataset IVA (data-3) and BCI Competition IV dataset II (data-4)). It was shown that the ensemble-based random subspace k-NN approach achieved the superior classification accuracy (CA) of 99.21%, 93.19%, 93.57% and 90.32% for data-1, data-2, data-3 and data-4, respectively against other models evaluated, namely linear discriminant analysis, support vector machine, random forest, Naïve Bayes and the conventional k-NN. In comparison with other classification approaches reported in the recent studies, the proposed method enhanced the accuracy by 2.09% for data-1, 1.29% for data-2, 4.95% for data-3 and 5.71% for data4, respectively. Moreover, it is worth highlighting that the RF feature selection technique employed in the present study was able to significantly reduce the feature dimension without compromising the overall CA. The outcome from the present study implies that the proposed method may significantly enhance the accuracy of MI data classification. How to cite this article Rashid M, Bari BS, Hasan MJ, Razman MAM, Musa RM, Ab Nasir AF, P.P. Abdul Majeed A. 2021. The classification of motor imagery response: an accuracy enhancement through the ensemble of random subspace k-NN. PeerJ Comput. Sci. 7:e374 DOI 10.7717/peerj-cs.374 Subjects Human-Computer Interaction, Artificial Intelligence, Brain-Computer Interface, Data Mining and Machine Learning Keywords Electroencephalography (EEG), Brain-computer interface (BCI), Motor imagery, Random forest, Ensemble learning, Common spatial pattern (CSP) INTRODUCTION The process of communication and control in human beings is largely dependent upon the peripheral nerves and muscles. When a healthy individual intends to do something, signals from a specific part of the brain area are sent via the peripheral nerves system to the corresponding muscles, which in turn perform the intended task. Many neurological disorders, which include stroke of the brain, injury to the spinal cord, cerebral palsy, muscle dystrophies, multiple sclerosis and amyotrophic lateral sclerosis amongst others, may impair the regular communication pathways of the signals (Bamdad, Zarshenas & Auais, 2015). If such neural disorders affect individuals considerably, the individuals may partly or generally begin to lose their voluntary motor control. In such scenarios, the individual would not be able to interact by any other means of communication with its surroundings. Researchers are continuously working on a variety of assistive technologies to address these concerns, and it is worth noting that the brain-computer interface (BCI) approach is amongst them. In every BCI system, specific brain signals are converted into control commands for the purpose of handling particular assistive devices (Wolpaw et al., 2002). Amongst the popular BCI applications are mind-controlled wheelchairs, speller, environment control, robotic arm control, biometrics, and emotion recognition (Rashid et al., 2020c). In addition, the BCI technologies are currently being extended from the known traditionally related medical areas to non-medical applications such as virtual reality and games (Rashid et al., 2020c). Many invasive and non-invasive neuroimaging approaches have been employed to record brain activity. The widely used invasive neuroimaging approaches are intracortical neurone recording and electrocorticography (ECoG). Conversely, the non-invasive approaches are electroencephalography (EEG), single-neuron recordings, magnetoencephalography, functional magnetic resonance imaging, functional near infraRed and positron emission tomography (Rashid et al., 2020d). Based on the recent BCI research activities (Padfield et al., 2019; Guan, Zhao & Yang, 2019), it is evident that EEG and ECoG are the most efficient modalities, so far, for BCI systems. Thus, it is noted that the patterns of mental activity should be decoded in such a way that people can modulate and interpret their thinking in order to deal with a specific BCI technology (Nicolas-Alonso & Gomez-Gil, 2012). In BCI, these signals are considered as control signals, and the broadly utilised neurological control signals are the steady visual evoked potential (SSVEP), the slow cortical potentials (SCP), the potentials evoked by P300, and the signal for motor imagery. The P300 based BCIs demonstrate comparatively better bit rate without requiring much training process. Nevertheless, the severity of ailment may have a substantial impact on the performance of P300 based BCIs. Although plenty of studies claim that patients with LIS can handle a P300 based BCI for longer periods, the information transfer rate of Rashid et al. (2021), PeerJ Comput. Sci., DOI 10.7717/peerj-cs.374 2/31 such patients is still smaller than the healthy users recorded in almost al (...truncated)