Table 1.
Number of samples from each Long-Evan rat, and number of samples for each class.
Fig 1.
Geodesic distance configuration in graph construction, where x and y represent the row and column indices (e.g., electrode coordinates).
Fig 2.
Proposed model architecture.
Table 2.
Generalization performance (accuracy and macro-averaged F1 score) of baseline model and graph-based models with varying connectivity (Rats 2 and 10 are excluded from final mean ± standard deviation evaluation as they were used in the validation set for hyperparameter tuning).
Fig 3.
Graph connectivity with five neighbors, defined by geodesic distance.
Table 3.
Generalization performance (accuracy and macro-averaged F1 score) of ablation studies (Rats 2 and 10 are excluded from final mean ± standard deviation evaluation as they were used in the validation set for hyperparameter tuning).
Fig 4.
Heatmap of F1-scores across different distance decay parameters and number of neighbors.
Fig 5.
Comparison of graph connectivity definitions in ablation studies.
(a) Euclidean distance–based graph. (b) Randomly constructed graph.
Table 4.
Within-rat performance (accuracy and macro-averaged F1 score) of baseline model, graph-based models with optimal connectivity, as well as ablation studies.