Design Consideration And Feature Extraction For Unsupervised Ensemble Based EEG Artifacts Eradication And Classification

Abstract

There is several motion artefacts that might appear in electroencephalography (EEG) data during recording. There were many ensemble based multimode decomposition methods had been designed to eradicate these artifacts. First the true and artefact signals are classified using machine learning (ML) methods. Then paper design and assess the qualitative performance of different unsupervised ensemble algorithms employing EEMD, ICA, EEMD-ICA, as well as EEMD-CCA for the elimination of EEG artefacts. For assessment the district database is taken into account. Actual and artificially created artefact data are also included in the evaluation. The motion artifacts especially ocular Eye blinks (EOG) and muscular (MMG) artifacts are considered for the research. The major challenge is to eradicate high peak eye blinks artifacts. Thus paper proposed wavelet based de noising to improve the signal strength after eradication of the artifacts. The performance is assessed based on the combination of DWT with all these ensemble based methods. The ultimate contribution of the paper is to generate the novel feature set for identification and classification of the EEG artifacts within the EEG signal data. The statistical features with different methods are calculated including PSNR, RMSE, mean, standard deviation, peak to peal difference, the EEMD-CCA-DWT out performs.