In each case, nine of the subsets were used for model fitting (training set), after which the observations in the remaining subset were predicted (testing set) and the mean squared error (MSE) of prediction was calculated.
For all species except the oryx, models of climatic suitability fit to training data from historical ranges produced larger areas of predicted suitability in North America than models fit to training data from modern ranges.
If experimental data are available, the model can be fitted (trained) to the data.
This occurs when models fit the training data too well and are not capable of predicting new samples.
Larger zones of potential climatic suitability were predicted in North America from models fit to historical training data compared to those fit to modern training data for the Asian elephant, cheetah and lion (Figures 1A, 1B, 1C, 1D, 2A and 2B).
The final models, fit using a training subset of the data, were cross-validated against both the overall dataset and a holdout subset of the data.
In the native range, models fit to modern training data for the same three species performed poorly when evaluated with historical testing data (Table 3).
For the cheetah and lion, more extensive regions of the American southwest and southern Great Plains appeared to be suitable based purely on climatic factors, especially from models fit to historical training data (Figures 1C, 1D, 2A, 2B).
In these studies model performance was assessed using measures of model fit, i.e., measuring how well models fitted the training data or test data on the same landscape as the training data.
These learning models fit well within the training data, but their performance with testing data significantly degrades [31].
