Neuro-Geospatial Model Links EEG Affective States to Environmental Context

Researchers have investigated a novel approach to classify human affective states using electroencephalography (EEG) data, augmented by environmental context. The study addresses the challenge that EEG and environmental datasets are rarely co-registered at an individual level. Instead, it proposes using literature-informed environmental priors as an auxiliary geospatial modality. The methodology involves combining 30-channel EEG data from the EAV benchmark with environmental representations derived from various open-source datasets like OpenAQ, Sentinel-2, and OpenStreetMap for the city of Astana. A dual-tower neural architecture was employed, integrating EEG-Conformer representations with a graph-based environmental encoder. Since direct individual exposure data was absent, environmental context was treated as a prior. The multimodal model achieved a classification accuracy of 76.2%, a notable improvement over 67.4% for EEG data alone. Control experiments confirmed that the gain was not solely due to environmental information. While demonstrating technical feasibility, the study emphasizes that it does not establish a causal exposure-affect association but provides a valuable framework for future mobile EEG-environment studies with jointly collected data.