Fig. 3: Coefficients of Hyperacusis Classifier.
From: Sparse machine learning of resting-state fMRI reveals brain-wide dysconnectivity in hyperacusis
Average model coefficients assigned to FC variables – a sign matrix and b magnitude histograms shown for non-zero values only. Dark-colored histograms correspond to ‘consistently-weighted’ FC variables. c FC coefficients: mean and standard deviation across model repetitions, shown in log scale. Yellow dots: FC edges with a non-zero average coefficient. Green dots: ‘consistently-weighted’ FC edges, i.e., having non-zero coefficient across more than 70% of the independent training repetitions. Black dashed line: coefficient mean = coefficient standard deviation. d Correspondence between mean model coefficient sign (x-axis) and direction of FC effect size (Hedges’ g; y-axis) comparing HA with CTR. Each dot represents an FC edge. Legend: same as (c). Because the class variable was encoded as ‘0’ for HA and ‘1’ for CTR, positive coefficients corresponded to FC edges that showed decreased connectivity trend in HA compared to CTR, and vice-e-versa for negative coefficients.
