Harnessing machine learning to predict habitat suitability of medicinal plants: insights from Oliveria decumbens under environmental change

Abstract Accurate habitat suitability modeling is essential for the conservation and sustainable cultivation of medicinal plants, particularly under changing environmental conditions. This study evaluates the habitat suitability of Oliveria decumbens in Fars Province, Iran, using five machine learning algorithms—Naive Bayes, Quadratic Discriminant Analysis, Linear Discriminant Analysis, Generalized Additive Models, and eXtreme Gradient Boosting. A total of 168 occurrence points and 18 environmental predictors, including topographic, climatic, and edaphic factors, were analyzed. Feature selection using Boruta and LASSO methods identified annual rainfall, elevation, and soil silt content as the most influential variables. QDA and LDA models outperformed others in spatial cross-validation, with AUC scores of 0.90 and 0.89, respectively, demonstrating high predictive accuracy and low uncertainty. Suitability maps indicated that optimal habitats are concentrated in the northern and northwestern regions of the province. Despite the robustness of the models, limitations related to environmental extrapolation and landscape dynamics remain. The findings provide a data-driven foundation for conservation planning and sustainable cultivation of O. decumbens . Future research should incorporate climate and land-use change projections to improve long-term resilience. This study demonstrates how integrating machine learning with spatial data supports evidence-based decision-making in ecological modeling.