BFRCNet: addressing the class imbalance problem in the rapid serial visual presentation paradigm for decoding.

Objective.Imbalanced sample sizes in rapid serial visual presentation (RSVP) can substantially compromise the classification accuracy of electroencephalogram (EEG) analyses based on RSVP system.Approaches.We propose a balanced strategy-based network for feature representation, recombination, and classification for RSVP paradigm (BFRCNet), a specialized neural network architecture designed to enhance classification under imbalanced EEG data conditions. This architecture comprises three stages, in feature representation stage, a pyramid structure integrates multiscale spatiotemporal patterns while mimicking visual physiological mechanisms to enhance EEG feature extraction. The recombination stage incorporates anchor samples as auxiliary categories, transforming the imbalanced distribution into a balanced representation. The last classification stage leverages a novel focal loss function that integrates class and sample weights, thereby enhancing the reward for minority samples.Main results.BFRCNet demonstrated significant performance improvements in addressing class imbalance for RSVP tasks, achieving balanced accuracy scores of 89.53% on THU and 90.15% on CAS datasets. This represents a 3.04% improvement on THU and 3.27% on CAS, substantially outperforming current state-of-the-art methods.Significance.This method to handle imbalanced EEG data effectively improves classification performance in class-imbalanced BCI-paradigms.