Fig. 5From: Proportion-based normalizations outperform compositional data transformations in machine learning applicationsAccuracy vs accuracy plots show how lognorm provides better accuracy for random forest than non-proportion-based transformations. Points for each plot represent scores of random forest classifier (accuracy) and random forest regressor (r2). Lognorm of raw DADA2 counts tables performs favorably compared to filtered alr (A), clr (B), raw DADA2 (C), filtered IQTREE PhILR (D), filtered SILVA DADA2 counts table (E), filtered SILVA DADA2 PhILR (F), filtered UPGMA PhILR (G), SILVA DADA2 counts table (H), and SILVA DADA2 PhILR (I). Each metadata feature of each dataset is shown in KBack to article page