Resting-State EEG Complex-Network Descriptors for Predicting Mental Arithmetic Performance

Abstract

Interindividual differences in mental arithmetic performance may be partly constrained by pre-task (resting-state) functional organization measurable from EEG (electroencephalogram). We operationalize organization as a windowed Pearson correlation connectivity graph and construct compact descriptors by combining a discrete Fourier transform–conjugated adjacency representation with eigen-spectral summaries, thereby capturing both local coupling patterns and global network structure in a unified feature space. Using a strictly subject-level, balanced leave-pair-out evaluation on a curated subset of the PhysioNet EEGMAT cohort (n = 20, 10 low- vs 10 high-performance; labels from the dataset's behavioral count-quality annotation), the resting-state pipeline produced perfect separation of performance groups (20/20 correct), indicating that baseline connectivity alone may encode sufficiently discriminative information for between-subject stratification. A complementary task-state setting reached 95% accuracy (19/20 correct), consistent with the repository's short arithmetic segments and their associated temporal constraints. The analysis explicitly avoids within-subject leakage by enforcing subject-wise partitioning, and performance estimates are accompanied by exact confidence intervals appropriate for small-sample evaluation. Given the limited cohort size and electrode density, these results should be interpreted cautiously and require replication on larger samples and higher-density montages; nevertheless, they provide evidence that performance-relevant information is detectable in baseline connectivity structure under rigorous subject-level evaluation.

Received: 11 January 2026 | Revised: 6 March 2026 | Accepted: 9 May 2026

Conflicts of Interest

The author declares that he has no conflicts of interest to this work.

Data Availability Statement

The EEG recordings analyzed in this study are publicly available from PhysioNet (EEGMAT) [1, 2]. To facilitate reproducibility, the analysis scripts used for feature extraction and model evaluation are provided as supplementary code [22]; all subject identifiers and performance labels used for the balanced cohort are listed explicitly in Table 1, and no additional human data were generated.

Author Contribution Statement

Miguel Angel Vargas Cruz: Conceptualization, Methodology, Software, Validation, Formal analysis, Investigation, Data curation, Writing – original draft, Writing – review & editing, Visualization, Supervision, Project administration.