Predictive Modeling for Targeted Archaeological Survey of Arsacid Period Sites in the Iranian Borderland Region of the Araxes River Valley

Bibliographic Data

ID6432884
AuthorsJiayue Wang (0000-0001-5068-631X, University of Chicago, corresponding author), Mehrnoush Soroush (0000-0002-9420-7824, University of Chicago), Sepideh Maziar (0000-0002-2253-7680, University of Chicago), Emad Khazraee (0000-0002-9873-2781, University of Chicago)
Year2025
Volume33
Issue1
Publication date2025-08-25
Open AccessYes
TypeARTICLE
VenueJournal of Archaeological Method and Theory (JOURNAL)
Journal identifiersISSN: 1072-5369 • E-ISSN: 1573-7764
PublisherSpringer Science+Business Media (PUBLISHER • DE)
DOI10.1007/s10816-025-09733-5
OpenAlexW4413631074
LanguageEN
References cited44

We have applied predictive modeling to archaeological sites, datable to the Arsacid period (ca. 3rd c. BCE–3rd c. CE), in the southern side of the Araxes River valley, in modern day Iran. Araxes River valley is a contested borderland between Iran, Armenia, and Azerbaijan. Systematic fieldwork has been historically limited by geopolitical tensions. The only systematic survey in the region and two salvage surveys have yielded a large number of sites that could belong to the Arsacid period—with varying levels of certainty—a period often undermined by scholarly attention to the subsequent Sasanian period (ca. 3rd–7th c. CE). We use this data as a training dataset to specify areas with high probability of finding more Arsacid period sites for future targeted surveys. Our paper follows previous research which has emphasized using machine learning approaches for archaeological predictive modeling. Acknowledging the inevitable oversight of social factors involved in settlement distribution, our statistical models produce probability of finding Arsacid period sites while also offering some insight into the physical attributes of settlement positioning in the Arsacid period. We tested several common machine learning (ML) models and suggested the suitability of Random Forest models for archaeological predictive modeling. We have applied Principal Component Analysis (PCA) to ensure component independence in modeling. Finally, we have investigated various theories regarding working with a subselection of principal components (PCs). Our study recommends using all PCAs, especially when the number of predictors does not create computational challenges. We support the argument made by Kvamme (2023) that lower PCs are particularly relevant for archaeological predictive modeling, but, we disagree with using them without other PCs

Archaeology · Art · Geography · Period (music · River valley · Archaeological Research and Protection · Archaeology and ancient environmental studies · Archaeology and Historical Studies · Geology

Tools

Open DOIOpen Access