Fig. 4: Calibrated and uncalibrated r-statistic and RvE plots for random forest using synthetic data with noise added.
From: Calibration after bootstrap for accurate uncertainty quantification in regression models
Distributions of r-statistic values and RMS residual vs. \(\hat \sigma\) plots for random forest, using the synthetic dataset with varying amounts of noise added. Gaussian noise with mean zero and standard deviation equal to 0.1 (panels a and e), 0.2 (panels b and f), 0.3 (panels c and g), and 0.5 (panels d and h) times the standard deviation of the training set with no noise added. Both uncalibrated and calibrated \(\hat \sigma\) are shown. Markers which are not filled in represent bins with fewer than 30 points. Statistics for each plot are summarized in Table 1.
