**5.4 Model evaluation**

Generating an independent data set is necessary to evaluate the model accuracy. This data set can be obtained through new surveys on supplementary areas or dividing the original data set in two. One of the data sets will be used to generate the models (train data) and the other to test it (test data). Usually, it is suggested to divide the original data set randomly and without reposition in 70% of the data to generate the model, and 30% to test it (Fielding and Bell, 1997; Hirzel and Guisan, 2002). It is also possible to divide the data considering their spatial pattern (Peterson et al., 2008). The area under the receiver operating characteristic curve (AUC) is usually used to evaluate the models. Values of AUC range from 0.5 for models with no predictive ability, to 1.0 for models giving perfect predictions (Swets, 1988). Some authors discussed the use of AUC to evaluate the model accuracy (Peterson et al., 2008) but nowadays, it is the most important method to this end (but see other alternatives on Thuiller, 2003). It is also possible to check the model accuracy conducting new surveys on the areas suggested as potential areas of occurrence by modelling.
