EAA2020: Abstract
Abstract is part of session #464:
Title & Content
Title:
Difficulties tracing and interpreting patterns in compositional data of metal artefacts. Why are the more complex methods not always useful?
Content:
Element compositions and isotope signals of metal artefacts represent a nice example of what can be designated as large datasets, especially in terms of the substantial number of features. In archaeology, we usually want to get insights about the provenance of the artefacts, consistency of the studied assemblage in terms of raw materials and employed technology, the similarity with other available assemblages from the given period etc. In terms of machine learning, various unsupervised learning methods are of help here. Feature selection and dimensionality reduction are usually followed by the application of various clustering methods to find meaningful groups in the dataset.
In the case study on a hoard of various artefacts from the La Tène period (4th – 1st century BC), we would like to illustrate some of the difficulties we faced while analysing the data by different unsupervised learning methods. The use of more complex statistical methods is not always leading to better interpretability of the results in archaeology terms.
In our case, sticking to less complicated machine learning methods proved useful in interpreting the results both in archaeological and raw material provenance terms.
The data analysis is implemented in an R environment ensuring reproducibility of the analysis.
In the case study on a hoard of various artefacts from the La Tène period (4th – 1st century BC), we would like to illustrate some of the difficulties we faced while analysing the data by different unsupervised learning methods. The use of more complex statistical methods is not always leading to better interpretability of the results in archaeology terms.
In our case, sticking to less complicated machine learning methods proved useful in interpreting the results both in archaeological and raw material provenance terms.
The data analysis is implemented in an R environment ensuring reproducibility of the analysis.
Keywords:
chemical composition, lead isotopes, provenance analysis, unsupervised learning, iron age
Downloads:
authors
Main authors:
Petr Pajdla1
Co-author:
Alžběta Danielisová2
Daniel Bursák2
Ladislav Strnad3
Jakub Trubač3
Daniel Bursák2
Ladislav Strnad3
Jakub Trubač3
Affiliations:
1 Department of Archaeology and Museology, Masaryk University
2 Institute of Archaeology CAS, Prague, v.v.i.
3 Charles University, Prague
2 Institute of Archaeology CAS, Prague, v.v.i.
3 Charles University, Prague