Unsupervised Classification of Neolithic Pottery From the Northern Alpine Space Using t-SNE and HDBSCAN

Bibliographic Data

ID13077439
AuthorsMartin Hinz (0000-0002-9904-6548, University of Bern, corresponding author), Caroline Heitz (0000-0001-7188-6775, Christian-Albrechts-Universität zu Kiel)
Year2022
Volume8
Issue1
Pages1183-1217
Publication date2022-01-01
Open AccessYes
TypeARTICLE
VenueOpen Archaeology (JOURNAL)
Journal identifiersISSN: 2300-6560 • E-ISSN: 2300-6560
PublisherDe Gruyter Open (PUBLISHER • PL)
DOI10.1515/opar-2022-0274
OpenAlexW4312657705
LanguageEN
Citations received1
References cited18

Terms of “Neolithic cultures” are still used to describe spatial and temporal differences in pottery styles across central Europe. These terms date back to research periods when absolute dating methods were lacking and typological classification was used to establish chronologies. Those terms are charged with problematic, biasing notions of social configurations: cultural homogeneity, spatial boundedness, and immobility. In this article, we present an alternative approach to pottery classification by using ceramics from dendrochronologically and C14-dated sites of the 40th–38th c. BC located in the northern Alpine Foreland. The newly developed methodology uses a computational unsupervised classification based on profile shape and additional nominal characteristics using t-Distributed Stochastic Neighbour Embedding and Hierarchical Density-Based Spatial Clustering of Applications with Noise for cluster analyses. Its role in our project was to provide a quantitative, algorithm-based approach to classify large datasets of pottery while simultaneously account for a large number of variables. This enabled us to find similarity structures that would escape human cognitive capacities on which typological classification is based on. It formed one pilar of a mixed method research approach combining qualitative and quantitative methods of pottery classification. Our results show that the premises of cultural homogeneity are untenable but can be methodologically overcome by using the proposed classification approaches

Archaeology · Cluster analysis · Foreland basin · Geography · Hierarchical clustering · Homogeneity (statistics · Machine learning · Pottery · Archaeology and ancient environmental studies · Computer Science · Image Processing and 3D Reconstruction · Artificial Intelligence · Geology · Paleontology

Tools

Open DOIOpen Access