Fig. 4: Performance of one-versus-one classification.

a Schematic illustrating the one-versus-one classification problem. b The mean area under the precision-recall curve across all 100 splits for different binary classifiers. Dark gray, real data. Light gray, shuffled data. c Violin plot representing the number of brain regions selected via the Boruta procedure for inclusion in the regression model (n = 100 splits for each classifier). The data centre represents median values, bounds of the inset represent the 25th and 75th percentile. Whiskers extend to 0th and 100th percentiles, excluding outliers (points beyond median ±1.5 * interquartile range). d Heatmaps showing the fraction of splits when a cortical (left) or thalamic (right) region was included in the regression model. The regions are sorted based on usage in all classifiers. Regions that were included in <75% of the splits across all conditions are not shown.