Graph Force Learning from EEG Signals for Mental Workload Classification

Modeling electroencephalography signals using graphs has recently gained a significant attention for different data analysis tasks in neuroscience. A key challenge is identifying the most effective representation that incorporates both the structural positioning of the electrodes and the signal’s statistical features to distinguish between brain states. Most graph based classification frameworks through the engineering of graph structural features, under-explore the positions of the electrodes. This study introduces a novel EEG classification approach that leverages the spatial arrangement of electrodes and a learned graph structure to define interaction forces between them. The latter are then used as node features for a brain activity classification using classical classifiers such as Support-Vector Machines. The resulting original method achieves a promising performance level in mental workload classification compared to conventional approaches.