Evaluating the efficacy of regression and machine learning models to predict prehistoric land-use patterns

Predictive modeling in archaeology is critical to making informed land management decisions and answering key anthropological research questions. However, archaeological predictive modeling su#30;ers from several theoretical, empirical, and analytical problems. To address these shortcomings, we build on theory from behavioral ecology and statistical innovations from ecology. Here we apply four modeling approaches and evaluate their performance using a threshold-independent measure. Two are regression- based: generalized linear (GLM) and generalized additive (GAM) models. Two are machine-learning based: maximum entropy (MaxEnt) and random forests (RF). The results of this analysis establish a foundation for future applications of predictive models in archaeology.