Predictive modeling for neolithic settlements in the Lingnan Region, South China

Predictive modeling is an effective tool to extrapolate the distribution pattern of ancient settlements, and the adaptative strategies of ancient populations to ever-changing environment. In this article, we build predictive models for settlements of three Neolithic periods in the Lingnan region, South China, which primarily encompasses the two present-day provinces of Guangxi and Guangdong. We process the geographic and topographic data of 793 sites to establish Logistic models, which we test with classification accuracy, AUC, Kvamme’s Gain statistic, and 5-fold cross-validation. In the meantime, we incorporate the subsistence strategies and landform types of 245 sites to observe how ancient populations responded to the matrix of environmental, technological, and cultural development. The results show that the models are powerful for predicting the distribution of Neolithic settlements in the region. Furthermore, they show that the distribution patterns of sites differ widely in the two provinces of Guangxi and Guangdong as well as in the three periods of the Early, Middle, and Late Neolithic.