The results show that the error range of the indices in analysing mean areal rainfall and simulated runoff narrowed gradually with increasing number of rain gauges up to some threshold, and beyond which the model performance did not show considerable improvements.
When phylogenetic covariates were included, weed model performance improved to 25.7%, but noxious model performance did not improve.
In conclusion, our study found that the applicability domain may not exist for microarray based clinical genomic research, and that predictive model performance did not depend on a measurement of distance between a validation sample and the training set used to create the model.
Moreover, the model performance did not stem from chance correlations, as R0 test values became negative when 75% of the pGI50 values were randomized (Clark and Fox, 2004) (Supplementary Fig. S4).
Model performances do not drop significantly, which is a sign of robustness.
The first section of "Model selection criterion AIC and predictive performance" does not make sense to me, are the authors talking about AIC capability to select a model that would predict new subjects?
Ideally, the points would fall on the diagonal line from bottom left to top right, and while the model is not perfect, the predicted performance does correlate somewhat with the experimental data, r=0.57.
It is important, however, that our assessment of their performance does not rely on the model assumptions, because violations can impact model performance.
Adding a spatial autocovariate term to the LR model (autologistic model) reduced the spatial autocorrelation and improved model performance, but did not correct the misidentification of the dominant environmental determinant.
Whenever possible, it is more convenient to use the sensitivity analysis during the optimization, which permits a fast evaluation of the model performance without doing a completely new analysis.
