Non-Invasive BCI by using EMD and Machine Learning: A Metaverse Interaction Perspective

Abstract

People with disabilities struggle to perform specific tasks throughout their daily life. However, BCI systems are developed to assist people struggling with motor impairment by transforming their thoughts into action. Non-invasive BCI systems use electroencephalogram (EEG) to record brain activities. In this study, we segment the EEG signals and then break the segment down into a few intrinsic mode functions using oscillation mode decomposition. Then the intrinsic mode functions are mined for feature extraction. The features mined are processed by different machine learning algorithms for categorization. Among the different algorithms, K-NN yielded the best results with an overall average accuracy score of 95.48%. This approach can be used in future to develop the brain driven metaverse interactive solutions.