Combining quantitative approaches to differentiate between backed products from discoidal and Levallois reduction sequences
Backed flakes (core edge flakes and pseudo-Levallois points) represent special products of Middle Paleolithic centripetal flaking strategies. Their peculiarities are due to their roles as both a technological objective and in the management of core convexities to retain its geometric properties duri...
| Autores: | , , , |
|---|---|
| Tipo de recurso: | artículo |
| Fecha de publicación: | 2022 |
| País: | España |
| Institución: | Universidad Autónoma de Madrid |
| Repositorio: | Biblos-e Archivo. Repositorio Institucional de la UAM |
| Idioma: | inglés |
| OAI Identifier: | oai:repositorio.uam.es:10486/705435 |
| Acceso en línea: | http://hdl.handle.net/10486/705435 https://dx.doi.org/10.1016/j.jasrep.2022.103723 |
| Access Level: | acceso abierto |
| Palabra clave: | Deep learning Discoid Geometric morphometrics Levallois Lithic analysis Machine learning Arqueología Historia |
| Sumario: | Backed flakes (core edge flakes and pseudo-Levallois points) represent special products of Middle Paleolithic centripetal flaking strategies. Their peculiarities are due to their roles as both a technological objective and in the management of core convexities to retain its geometric properties during reduction. In Middle Paleolithic contexts, these backed implements are commonly produced during Levallois and discoidal reduction sequences. Backed products from Levallois and discoidal reduction sequences often show common geometric and morphological features that complicate their attribution to one of these methods. This study examines the identification of experimentally produced discoidal and recurrent centripetal Levallois backed products (including all stages of reduction) based on their morphological features. 3D geometric morphometrics are employed to quantify morphological variability among the experimental sample. Dimensionality reduction though principal component analysis is combined with 11 machine learning models for the identification of knapping methods. A supported vector machine with polynomial kernel has been identified as the best model (with a general accuracy of 0.76 and an area under the curve [AUC] of 0.8). This indicates that combining geometric morphometrics, principal component analysis, and machine learning models succeeds in capturing the morphological differences of backed products according to the knapping method |
|---|