Shifted and Weighted LFCC Features for Hand Movements Recognition Using EEG Signals

The Brain computer interface (BCI) technology attracts many researchers due to its vital applications in medicine and biomedical domains. Decoding the motor imagery electroencephalography signals (MI-EEG) to find the corresponding brain activities constitutes the core of the BCI system. Researchers use both classical machine learning and deep learning to identify such activities. In this work, left/right hand movement is recognized using MI-EEG signals features. The EEG signal features vector is derived from the wavelet transformation (WT) and the linear frequency cepstral coefficient (LFCC) to feed an SVM classifier. Two publicly available MI-EEG datasets are tested, BCI competition III b and BCI competition IV 2b. Firstly, the classification is performed using all features without any modification for the LFCC features. Secondly, the LFCC features are shifted and weighted (SWLFCC) by genetic algorithm (GA) to improve the classification accuracy. The average classification accuracy was improved and achieved 92.19 and 87.23% for BCI competition III b and BCI competition IV 2b datasets respectively. Compared to the works using the same datasets, the performance to identify left and right hand is improved by 2.4% for the BCI competition III b. There was very small improvement for the second dataset though the deep learning was used for the last best accuracy. The novel SWLFCC features showed their efficiency in BCI system to identify MI left/right hand movement. The SWLFCC features are tuned by GA optimization to adapt any nonlinearity existing between tasks and MI-EEG signals.