Figure 6.

Prediction errors [Weighted root mean squared error (wRMSE)] of individual regression models and the proposed ensemble model for (A) bladder (B) rectum, in the presence of simulated dosimetric outliers (see text). For bladder prediction, the proposed ensemble method predicts significantly better than stepwise (p < 0.001), ridge (p < 0.001), and performs similarly well as lasso (p = 0.753) and elastic net (p = 0.841). For rectum prediction, the proposed ensemble method predicts significantly better than stepwise (p < 0.001) and ridge (p < 0.001), and performs similarly well as lasso (p = 0.365) and elastic net (p = 0.373).