Towards automated classification of cognitive states: Riemannian geometry and spectral embedding in EEG data

Abstract

Our research explores the application of Riemannian geometry and spectral embedding in the context of electroencephalogram (EEG) signal analysis for cognitive state classification. Leveraging the PyRiemann library and the AlphaWaves dataset, our study employs covariance estimation and the minimum distance to mean (MDM) classifier within a machine learning pipeline. The classification accuracy is assessed through stratified k-fold cross-validation. Furthermore, we introduce a novel visualization approach by calculating the spectral embedding of covariance matrices, providing insights into the underlying structure of the EEG epochs. Our findings showcase the potential of Riemannian geometry and spectral embedding as powerful tools in the domain of EEG-based cognitive state classification, contributing to the broader field of brain signal analysis and paving the way for automated and advanced neurocognitive studies.