Modelling Local Strontium Isotope Baselines from Human Isotopic Data Using Machine Learning: Implications for Past Mobility Studies

Estimating bioavailable strontium isotope baselines is a critical step in archaeological mobility studies. Conventional proxy-based approaches, which rely on environmental samples, such as soils, fauna, plants, or waters, are not always feasible and may be affected by environmental, taphonomic, or sampling-related biases. Here, we introduce MobSr, a machine-learning-based model designed to estimate probable local 87 Sr/ 86 Sr baselines directly from human enamel isotope data. The model combines four first-layer classifiers—Naïve Bayes, k-nearest neighbours, support vector machine, and Random Forest—with a Bayesian ensemble layer that integrates classifier outputs and refines local/non-local probability estimates. MobSr was trained on a dataset of approximately 3000 human enamel samples from 86 archaeological sites, using published expert-derived local/non-local classifications. Local isotopic ranges are estimated by analysing posterior probabilities assigned to individual observations. Model performance was evaluated through general cross-validation and through site-specific case studies, including a La Tène period cemetery from the Czech Republic, an Early Medieval cemetery from Finland, and multi-period burial assemblages from Bahrain in the Arabian Peninsula. The results show 87–95% agreement with published local/non-local classifications, depending on the evaluation strategy and treatment of atypical cases. These values quantify concordance with the source-study labels. These findings demonstrate that MobSr can replicate and refine local baseline estimates without requiring additional environmental sampling. The model offers an efficient tool for preliminary mobility assessment and provides a complementary strategy to traditional baseline construction in bioarchaeological research. In addition, the use of a large training dataset enables broader exploration of expert-derived locality assignments across predominantly sedentary populations from different regions and periods.