Skip to main content
. Author manuscript; available in PMC: 2018 Jan 15.
Published in final edited form as: Neuroimage. 2016 Feb 23;145(Pt B):346–364. doi: 10.1016/j.neuroimage.2016.02.041

Figure 4.

Figure 4

Simulated data results: (a) Cross-validated AUC for HYDRA (left) and K-means/SVM (right) binary classification. (b) Cross-validated ARI for the clustering result of HYDRA (left) and K-means (right). The results are reported for different values of the parameter K. Error bars are centered around the mean and indicate variance. Both the classification accuracy and the cluster stability were maximized at K = 3 for HYDRA, agreeing with the intrinsic dimensionality of the heterogeneous group. The classification accuracy obtained by K-means/SVM remained relatively stable for different values of K. However, the clustering stability was maximized for K = 2, demonstrating that higher reproducibility does not necessarily imply successful heterogeneity detection.