Machine Learning Based Typology Development in Archaeology

摘要

Formalizing and objectifying the process of artefact classification is an old wish of many archaeologists. On the other hand, data mining in general and machine learning in particular have already inspired many disciplines to introduce new paradigms of data analysis and knowledge discovery. Hence, this article aims for reviving the Typological Debate by adapting approved methods from other fields of science to archaeological data. To this end, we extensively discuss the concept of similarity and assess the suitability of machine learning techniques for the purposes of classification and typology development. Our methodology covers all steps starting from unordered, unlabelled objects to the emergence of a consistent and reusable typology. The application of this process is exemplarily illustrated by classifying the vessels from a Late Bronze Age cemetery in Eastern Saxony. Despite the individual character of these vessels, we achieved class prediction rates of more than 95%. Such a success was only possible, because we permanently reconciled the output of the learning algorithms with our own expectations in order to identify and eliminate the systematic errors within the typology.