Figure 4
Effect of individual trees (ntree) on model performance and computation time for Lemna minor, Spirodela polyrhiza and Nuphar lutea. Along an increase in number of individual trees, no clear increase in kappa values can be observed, while a clear increase in computation time is present. Performance values were calculated for a random forest trained with data of which 18% was missing, without outlier and correlated variable removal, followed by median imputation. During model training, 5-fold cross-validation was applied and repeated with ten different runs, for which the average computation time was determined.
