EEG-based cognition-aware task classification and scheduling using enhanced fuzzy transition modeling

The cognitive state modeling (CSM) problem is typically formulated as a classification problem, limiting the application of the CSM for adaptive real world applications, where the desired outputs are cognitive states to be desired and the inferred ones have to be used for decision making. While conventional methods classify states of the brain, they have not yet been able to connect the class to the task level. To address this, this paper suggests a neurosymbolic model of cognition as a continual latent process rather than an incremental labelling process. It includes a Pseudo Task based Neural State Encoder (PNSE) to encode EEG windows into a structured hyperspherical embedding space, a Neural Transition Graph Network (NTGN) to learn the relationships between cognitive states and tasks, and a Temporal Pseudo-Task Boundary Model (TPBM) to capture the temporal evolution of cognitive states. The neurosymbolic decision layer is used to produce a single scheduling metric using a neural compatibility score, a probabilistic transition measure and a symbolic fuzzy membership, while a fuzzy inference engine is used to categorize candidate task classes with fuzzy membership grades. The framework was tested on a multi-session multi-task EEG cognitive dataset (COG-BCI) using a protocol that was subject independent. The Silhouette Score, Hit Rate, Normalized Discounted Cumulative Gain (NDCG) and Mean Reciprocal Rank (MRR) achieved experimental results of 73.7%, 71.43%, 91.58% and 76.67%, respectively, in the fuzzy membership space. Moreover, the proposed system had a precision of 81.1%, a recall of 83.4% and achieved an accuracy of 83.47% and an F1 score of 82.7%. The outcomes illustrate the possibility of getting cognitive modelling from EEG data to enable active recognition of cognitive states, and the inference and scheduling of uncertain tasks. The proposed framework provides a tractable, temporally unified and cognitively flexible foundation for future decision-support systems that would benefit from both the interpretability and adaptability of neural representation learning and symbolic reasoning and temporal modelling.