Advancing predictive modeling in archaeology: An evaluation of regression and machine learning methods on the Grand Staircase-Escalante National Monument

Abstract

Predictive models are central to both archaeological research and cultural resource management. Yet, archaeological applications of predictive models are often insufficient due to small training data sets, inadequate statistical techniques, and a lack of theoretical insight to explain the responses of past land use to predictor variables. Here we address these critiques and evaluate the predictive power of four statistical approaches widely used in ecological modeling-generalized linear models, generalized additive models, maximum entropy, and random forests-to predict the locations of Formative Period (2100-650 BP) archaeological sites in the Grand Staircase-Escalante National Monument. We assess each modeling approach using a threshold-independent measure, the area under the curve (AUC), and threshold-dependent measures, like the true skill statistic. We find that the majority of the modeling approaches struggle with archaeological datasets due to the frequent lack of true-absence locations, which violates model assumptions of generalized linear models, generalized additive models, and random forests, as well as measures of their predictive power (AUC). Maximum entropy is the only method tested here which is capable of utilizing pseudo-absence points (inferred absence data based on known presence data) and controlling for a non-representative sampling of the landscape, thus making maximum entropy the best modeling approach for common archaeological data when the goal is prediction. Regression-based approaches may be more applicable when prediction is not the goal, given their grounding in well-established statistical theory. Random forests, while the most powerful, is not applicable to archaeological data except in the rare case where trueabsence data exist. Our results have significant implications for the application of predictive models by archaeologists for research and conservation purposes and highlight the importance of understanding model assumptions.

Theoretical Repositioning of Automated Remote Sensing Archaeology: Shifting from Features to Ephemeral Landscapes

“…Indeed, such vegetative proxies have been incorporated into many predictive modeling efforts to locate archaeological sites (e.g., Agapiou 2020; Calleja et al 2018; Davis et al 2020a;Hegyi et al 2020;Kirk, Thompson and Lippitt 2016;Yaworsky and Codding 2018). In several instances, vegetative indices have been used in conjunction with other automated remote sensing methods (e.g., Davis et al 2020a;Kirk, Thompson and Lippitt 2016;Yaworsky et al 2020).…”

“…While this method is most explicitly rooted in statistical theory, the predictive modeling approach itself is entrenched in cultural ecology approaches that have been part of standard archaeological frameworks for nearly a century (Butzer 1982;Steward 1937Steward , 1955. Very similar approaches persist in archaeology today, many of which rely on at least some remotely sensed environmental information (e.g., Davis, DiNapoli and Douglass 2020;Verhagen and Whitley 2012;Yaworsky et al 2020).…”

A predictive model for the ichnological suitability of the Jezero crater, Mars: searching for fossilized traces of life-substrate interactions in the 2020 Rover Mission Landing Site

“…As pointed out by other authors (Brandt, Groenewoudt & Kvamme, 1992), one weakness of the weighted approach to predictive modeling is the element of subjectivity in the weights assigned to each suitability factor. Other statistical approaches are available to deal with the subjectivity factor, e.g., Bayesian statistics (Millard, 2005), machine learning (Yaworsky et al, 2020) and graph similarity analysis (Mertel, Ondrejka & Šabatová, 2018) have been applied to predictive modelling. However, there are no field data for the Jezero crater and only Earth-type life is known, therefore, application of more sophisticate statistical methods would bring the risk of overinterpretation, that is, reading too much into the limited dataset available.…”

Does the Locally-Adaptive Model of Archaeological Potential (LAMAP) work for hunter-gatherer sites? A test using data from the Tanana Valley, Alaska

“…Predicting archaeological potential has been an active area of methodological research for a long time [ 11 , 12 ] and a range of statistical methods have become established in the literature [ 13 – 15 ]. Most methods involve regression models designed to estimate the probability of finding archaeological material given known site locations and a set of predictor variables.…”

“…Consequently, any regression model applied to such data is very likely to have trouble distinguishing high and low potential locations. Recently, the impact of this flaw has been demonstrated quantitatively [ 15 ].…”