Fig. 3.

Values of regression coefficients in A) Lasso and B) Elastic Net regressions, respectively, where the bars' height corresponds to the coefficients' values for selected variables. Coefficients of all other variables are shrunk to zero (not shown) by the regressions. Variable importance in C) Gradient Boosting (GBoost) and D) Random Forest (RF) regressions, with the bars' height corresponding to estimated importance. Only variables with P < 0.1 (according to either Pearson, Kendall, or Spearman correlations with R₀) are included in GBoost and RF regressions. MSE values are scaled to the constant value model and averaged over 200 cross-validations. P-values correspond to the statistical significance of obtained MSE's compared to the baseline model. Variable names are indicated on the horizontal axis.