Accurate Predictive Modeling of Conservation Status in Animal Species Using Supervised Learning

Conservation management to mitigate extinction of wildlife becomes more crucial than ever as global impacts due to anthropogenic activities and climate change continue to create devastation for species around the globe. The International Union of Conservation of Nature (IUCN) RedList does not currently utilize genetic information to assess species conservation status despite the availability of molecular data. Here we use over 7300 animal studies collated from the MacroPopGen database, and over 450 published articles from the public repository DataDryad, focused on conservation and population genetics, sampling across a variety of invertebrate and vertebrate taxa, and using IUCN classifications to predict species endangerment across and within animal taxa using machine learning. We test hypotheses and show significant (p < 0.05) (a) decreased genetic diversity, and (b) increased genetic differentiation in bird and fish taxa with increased degree of endangerment. Additionally, our models were able to accurately (overall accuracy of 93.16%) predict species threat levels classified by the IUCN using both measures of genetic diversity and differentiation with IUCN assessment criteria. We propose that future studies that assess conservation status of animal taxa utilize a combination of predictor variables, that include available genomic data, along with demographic, phenotypic, and census data.