Machine learning applications in archaeological practices: a review

TL;DR

Problem: limited understanding of how ML methods are applied across archaeological subfields and the methodological pitfalls that accompany their use. Approach: a rapid systematic review of 135 publications (1997–2022) with a broad taxonomy of ML families, archaeological subfields, inputs, and evaluation metrics, plus a proposed workflow for coherent, transparent practice. Findings: ANNs and ensemble methods dominate, with automatic structure detection and artefact classification as the most common tasks; ML uptake varies by subfield and data type, and methodological issues—especially around reporting, data availability, and interpretability—are repeatedly highlighted. Significance: the study provides a practical framework and best-practice recommendations to improve rigor, reproducibility, and collaboration in applying ML to archaeology, while acknowledging limitations and ethical considerations.

Abstract

Artificial intelligence and machine learning applications in archaeology have increased significantly in recent years, and these now span all subfields, geographical regions, and time periods. The prevalence and success of these applications have remained largely unexamined, as recent reviews on the use of machine learning in archaeology have only focused only on specific subfields of archaeology. Our review examined an exhaustive corpus of 135 articles published between 1997 and 2022. We observed a significant increase in the number of publications from 2019 onwards. Automatic structure detection and artefact classification were the most represented tasks in the articles reviewed, followed by taphonomy, and archaeological predictive modelling. From the review, clustering and unsupervised methods were underrepresented compared to supervised models. Artificial neural networks and ensemble learning account for two thirds of the total number of models used. However, if machine learning models are gaining in popularity they remain subject to misunderstanding. We observed, in some cases, poorly defined requirements and caveats of the machine learning methods used. Furthermore, the goals and the needs of machine learning applications for archaeological purposes are in some cases unclear or poorly expressed. To address this, we proposed a workflow guide for archaeologists to develop coherent and consistent methodologies adapted to their research questions, project scale and data. As in many other areas, machine learning is rapidly becoming an important tool in archaeological research and practice, useful for the analyses of large and multivariate data, although not without limitations. This review highlights the importance of well-defined and well-reported structured methodologies and collaborative practices to maximise the potential of applications of machine learning methods in archaeology.

Figures (10)

• Figure 1: Review process from source selection to analysis. Inspired by the PRISMA 2020 flow diagram page_prisma_2021. Reason 1 = Ineligible with automation tool; Reason 2 = Non-English record; Reason 3 = Full text not accessible; Reason 4 = Non-journal-based publications; Reason 5 = Absence of abstract; Reason 6 = Archaeology and machine learning keywords from the list not present in the text; Reason 7 = Archaeology and machine learning keywords from the list are not present in the abstract or in the title; Reason 8 = Preliminary exclusion (i.e. no access to publication, publications or contribution by current authors, entire books, non-academic reports, preprints, potentially predatory journal; Reason 9 = Excluded based on the title; Reason 10 = Excluded based on abstract; Reason 11 = Excluded based on the full text first reading; Reason 12 = Full text does not involve archaeological research; Reason = 13 Full text does not involve machine learning methods as defined in our protocol; Reason 14 = Conflicts of interest (publication by the authors of this review or in which the authors contributed); Reason 15 = Theory or review paper. Figure created using Microsoft Word and Inkscape.
• Figure 2: The fourth field of information recorded in the review presents significant characteristics to explain variation in machine learning applications in archaeology and their related classes/categories. One study case might have been attributed to several subfields or architecture categories. Figure generated with R 4.2.2 (code available in supplementary material 3) and additional editing with Inkscape.
• Figure 3: Number of publications per year between 1997 and 2022, in light blue the articles published after 2018 concentrated more than 80% of the publications. The dashed line represents publications from 1 January 2023 to 31 September 2024. Figure generated with R 4.2.2 (code available in supplementary material 3) with additional editing in Inkscape.
• Figure 4: Number of articles published per country based on the country of the first author’s affiliation. Figure generated with R 4.2.2 (code available in supplementary material 3).
• Figure 5: (A) Number of articles from each archaeological subfield between 1997 and 2023. (B) Number of articles from each architecture class between 1997 and 2023. Empty bar charts represent the number 1. Figure generated with R 4.2.2 (code available in supplementary material 3).