Predictive modeling in geoarchaeology: An evaluation of machine learning algorithms and topographic variables on the Serranópolis City - Brazil

Abstract

The search for predictive models in archaeology using geographical data and artificial intelligence (AI) has made progress, but there is a scarcity of studies focusing on South America. Serranópolis, in Goiás, Central-West Brazil, emerges as an ideal landscapefor applying AI and geographical data to predict archaeological sites. This study aimed to predict potential archaeological locations in Serranópolis using topographic variables and four supervised classification algorithms: C5.0, Random Forest (RF), eXtreme Gradient Boosting (XGBoost), and Gradient Boosting Machine (GBM). The methodology involved sample points representing open-air archaeological sites and rock shelters, along with randomly selected pseudo-absence samples. Eighteen topographic variables from a digital elevation model were considered. RF outperformed other algorithms, producing a probability map of site occurrence. Key predictors included sky view factor, roughness, topographic depression, and slope. The findings enhance our understanding of archaeological potential in this region, highlighting the effectiveness of RF in predictive modeling.