Contributed Talk Sessions | Poster Sessions | All Posters | Search Papers

Poster Session A: Tuesday, August 12, 1:30 – 4:30 pm, de Brug & E‑Hall

Graph-Based Learning for EEG Workload Classification: Eliminating the Need for Calibration

Ernesto Bocini¹, Iris Kremer, Pablo Mainar Jovani; ¹EPFL - EPF Lausanne

Presenter: Ernesto Bocini

Overwork and mismanagement of brain workload are leading causes of mental distress. Monitoring Mental Workload (MWL) is therefore crucial for personalized wellbeing recommendations. Electroencephalography (EEG) signals have been shown to assess cognitive states during specific tasks effectively. However, modern methods depend heavily on signal pre-processing and hand-crafted features, limiting their ability to generalize to unseen subjects and often requiring calibration. This study investigates the usage of two graph-based deep learning approaches to tackle these problems. They are tested on two datasets alongside other widely used EEG classifiers. The leave-one-subject-out cross-validation (LOSOCV) strategy is used to tackle the cross-subject generalization problem frequently encountered when using EEG. The results show that models leveraging the graph structure of the EEG data consistently outperform comparison methods on both datasets, achieving strong performance without calibration to new subjects. These results highlight the potential of graph-based approaches as a foundation for future improvements in real-time mental health monitoring and personalized interventions.

Topic Area: Methods & Computational Tools

Extended Abstract: Full Text PDF