Evaluating Neurostellar’s Wearable Device Along with Its Robust Multimodal Approach for Real-Time Mental State Assessment: A Validation Study

Consumer wearables for mental state assessment face scepticism regarding data quality. This study evaluated whether a consumer-grade EEG/PPG wearable headband (Neurostellar’s wearable device) can reliably capture mental states in real-world settings. We compared the wearable device's EEG (AF7, AF8) and PPG-derived heart rate variability (HRV) signals with a laboratory-standard EEG/ECG device across 31 resting sessions (15 adults), assessing univariate correlations across 18 EEG, 12 HRV features. Participants then completed four cognitive and affective tasks while using the wearable device, providing subjective feedback. Novel multimodal features, combining short-window individual metrics with long-window complexity from multi-feature cluster sequences, were utilised in machine learning (Random Forest, Gradient Boosting) to classify tasks and predict changes in subjective relaxation and focus. Neurostellar’s wearable device showed moderate-to-strong correlations with lab measurements for numerous EEG (eyes-closed: 12; eyes-open: 14) and HRV (eyes-closed: 9; eyes-open: 6) features in >50% of recordings. The advanced multimodal features significantly classified all tasks (Random Forest: z=13.08-13.11; Gradient Boosting: z=13.00-13.16, all p<0.001) and predicted combined relaxation/focus ratings (Random Forest: z=2.37; Gradient Boosting: z=1.79, all p<0.05) versus chance. These findings suggest that consumer wearables, employing robust multimodal feature engineering, can effectively capture psychophysiological changes and track mental states, such as focus and relaxation, in naturalistic environments, offering potential for real-world neuroscience applications. Limitations include a modest sample size and restricted demographic diversity, which warrant further research.