EEG Motion Artefact Classification and Removal Using SVM and Optimum Reduce Order Filter (OROF) Design

Abstract

Electroencephalography (EEG) data often contain motion artefacts during acquisition, making it essential to remove them early in the analysis of neurological disorders. This paper presents a machine learning (ML)-based approach for detecting and classifying motion artefacts in EEG data. Two distinct databases, including original and synthetically generated artefact data, are utilized for evaluation. The classification process employs statistical features extracted from the EEG motion artefact database, which are then tested using ML classifiers to determine accuracy. Among the tested classifiers, cubic support vector machine (SVM) demonstrates the highest classification accuracy and computational efficiency.Once artefacts are identified, an optimal reduced-order filter (OROF) is proposed for artefact removal. The filter design is initially validated using an infinite impulse response (IIR) filter, followed by min-max optimization to ensure the integrity of the true EEG signals. The effectiveness of the proposed filter is assessed using a multichannel EEG artefact dataset. Finally, the peak signal-to-noise ratio (PSNR) is evaluated to verify the filter’s performance in preserving EEG signal quality.The proposed approach successfully enhances EEG signal processing by accurately classifying motion artefacts and efficiently filtering them while maintaining signal integrity.