@article{fdi:010073978,
  title = {{W}ithout quality presence-absence data, discrimination metrics such as {TSS} can be misleading measures of model performance},
  author = {{L}eroy, {B}. and {D}elsol, {R}. and {H}ugueny, {B}ernard and {M}eynard, {C}. {N}. and {B}arhoumi, {C}. and {B}arbet-{M}assin, {M}. and {B}ellard, {C}.},
  editor = {},
  language = {{ENG}},
  abstract = {{T}he discriminating capacity (i.e. ability to correctly classify presences and absences) of species distribution models ({SDM}s) is commonly evaluated with metrics such as the area under the receiving operating characteristic curve ({AUC}), the {K}appa statistic and the true skill statistic ({TSS}). {AUC} and {K}appa have been repeatedly criticized, but {TSS} has fared relatively well since its introduction, mainly because it has been considered as independent of prevalence. {I}n addition, discrimination metrics have been contested because they should be calculated on presence-absence data, but are often used on presence-only or presence-background data. {H}ere, we investigate {TSS} and an alternative set of metricssimilarity indices, also known as {F}-measures. {W}e first show that even in ideal conditions (i.e. perfectly random presence-absence sampling), {TSS} can be misleading because of its dependence on prevalence, whereas similarity/{F}-measures provide adequate estimations of model discrimination capacity. {S}econd, we show that in real-world situations where sample prevalence is different from true species prevalence (i.e. biased sampling or presence-pseudoabsence), no discrimination capacity metric provides adequate estimation of model discrimination capacity, including metrics specifically designed for modelling with presence-pseudoabsence data. {O}ur conclusions are twofold. {F}irst, they unequivocally impel {SDM} users to understand the potential shortcomings of discrimination metrics when quality presence-absence data are lacking, and we recommend obtaining such data. {S}econd, in the specific case of virtual species, which are increasingly used to develop and test {SDM} methodologies, we strongly recommend the use of similarity/{F}-measures, which were not biased by prevalence, contrary to {TSS}.},
  keywords = {{AUC} ; ecological niche models ; model evaluation ; prevalence ; species ; distribution models},
  booktitle = {},
  journal = {{J}ournal of {B}iogeography},
  volume = {45},
  numero = {9},
  pages = {1994--2002},
  ISSN = {0305-0270},
  year = {2018},
  DOI = {10.1111/jbi.13402},
  URL = {https://www.documentation.ird.fr/hor/fdi:010073978},
}