Integrating GIS-MCDA and Machine Learning for Geoarchaeological Habitation Suitability Modeling in the Karst Mountain Environment of Biokovo Nature Park

Mountain karst landscapes present major challenges for reconstructing past habitation because suitable environments are spatially fragmented and archaeological remains are often dispersed. This study integrates GIS-based multicriteria decision analysis (GIS-MCDA) and machine learning methods to model habitation suitability in Biokovo Nature Park, Croatia. Twenty-one morphometric, hydrogeomorphological, climatic, and archaeological criteria, together with 804 reference polygons, were used to develop four habitation suitability models: Equal-Weight GIS-MCDA, Analytic Hierarchy Process (AHP) GIS-MCDA, Random Forest (RF), and XGBoost. All models showed strong discrimination on the held-out validation dataset, with AUC values of 0.9152 for Equal Weight, 0.9447 for AHP, 0.9876 for RF, and 0.9886 for XGBoost, although these values should be interpreted within the adopted reference-sample design. The final XGBoost model indicates that favourable habitation environments are limited and spatially discontinuous, with 16.72% of the modelled area falling within the two highest Jenks classes used for cartographic interpretation. Suitable zones are concentrated within distinct karst micro-landscapes, particularly around dolines. The results support the interpretation of Biokovo as a selectively used mountain landscape and demonstrate the potential of combining GIS-MCDA and machine learning approaches for archaeological prospection in Mediterranean and Dinaric karst environments.