Figure 3
From: Estimation of model accuracy by a unique set of features and tree-based regressor
Filtering out outlier targets from the training set reduces the error in quality estimation (A), and does not affect the identification of the best models (B). The box plots depict the results of Leave-One-Target-Out experiments with and without data filtering in the training set. (A) The median of the RMSE is significantly reduced by data filtering (Wilcoxon one-sided test, with a p-value of 0.005). (B) Data filtering does not affect the distribution of LOSS (the quality differences between the top-ranking model, and the best model in the set). Many of the worse performing outliers in the plots are proteins that were filtered out from the training set.
