Fig. 4

Measuring %parent offers improvements in patient classification performance. a Comparative boxplots (lower panel) of balanced accuracy rates for each dataset and feature set: %total, %parent using average-linkage hierarchical clustering, and %parent using HOPACH. Values plotted are from a 5-fold CV with 20 repetitions, averaged across each repetition. The rank of each feature set within each dataset is shown in the bubble plot (upper panel), with rank 1 being the best (highest mean / lowest variance) and rank 3 being the worst (lowest mean / highest variance)