FIG. 3.

ROI participation in most informative features, separated by network (from Power et al., 2011). CRF values are the average across all runs after dimension reduction resulted in peak accuracy (308, 400, and 627 features). RF values are taken from Chen and colleagues (2015) in a model containing 100 features. (A) Percentage of ROIs from each network. (B) Normalized percentages (adjusted for the variable numbers of ROIs per network among the total of 24,090 connections, before dimension reduction). CRF, conditional random forest. Color images available online at www.liebertpub.com/brain