Stronger Together: Modeling species distribution under limited occurrence data with Deep Learning and Synthetic Oversample

Species Distribution Models face a major challenge in dealing with rare or under-sampled species, due to statistical biases resulting from the lack of training data. To address this issue, in this study we propose the use of the MLSMOTE technique (Multi-Label Synthetic Minority Over-sampling Technique) to generate synthetic samples of minority classes within the most imbalanced labels of the dataset. To build the dataset, we used presence records of 48 tetrapod species endemic to the Caatinga Domain, which also served as the study area for our analyses.