Enhancing the Performance of Event-Related Potential-Based Brain–Computer Interfaces under Cognitive Distraction: A Multiwindow Adaptive Approach

Event-related potential (ERP)-based brain- computer interfaces (BCIs) require focused attention to presented stimuli. However, their applications in real life frequently involve environments that demand multitasking and impose cognitive distraction. Such distractions degrade ERP amplitudes and consequently reduce BCI performance. This study proposes a multiwindow adaptive model to mitigate the adverse effects of cognitive distraction on visual ERP-based BCIs. The proposed approach divides poststimulus intervals into multiple overlapping windows, each with dedicated spatial filters and classifiers that continuously update through adaptive semi-supervised learning. Offline experiments on a BCI control dataset collected during concurrent speaking demonstrate that the proposed method significantly outperforms single-window or fixed (i.e., nonadaptive) models, yielding an accuracy of 91.08%. Further validation in an online experiment confirms that the multiwindow adaptive approach effectively restores BCI performance, achieving an accuracy of 93.20% despite cognitive distraction. These findings highlight the practical benefits of temporally tailored feature extraction and continuous adaptation for real-world ERP-based BCIs, enabling robust performance even under cognitive distraction.