Fig. 3From: Proportion-based normalizations outperform compositional data transformations in machine learning applicationsTypical box and whiskers plots of random forest scores for single features and selected transformations. This plot shows the scores of the random forest regressions and classifiers on BMI (A) and “stool vs swab” (B) in the Jones dataset, respectively. The x-axis shows each transformation and the y-axis shows the score, which was r2 for the quantitative BMI and accuracy for the categorical stool vs swab (type). For the sake of simplicity, only the transformations made from the unfiltered counts table are shown and the Silva/LTP PhILR transformations are shown as the lone representative of the various PhILR transformations we examinedBack to article page