Statistical learning methods applied to stratigraphic and pottery data: An aid to establishing archaeological periodisation

Chronology, and hence periodisation of archaeological sites according to the major transformations that affect them, is an essential prerequisite for any historical discourse. The main sources that can be used to establish this periodisation, the temporality of archaeological sites, are (i) stratigraphy, which corresponds to the succession of anthropic levels at the origin of the construction of a relative chronology, and (ii) archaeological material, and more specifically pottery that is omnipresent in excavations, with a typology that evolves rapidly over time; these two factors make pottery a highly valuable chronological source. This work presents an original interdisciplinary approach in which archaeology and statistics work together to produce a periodisation using pottery and stratigraphy. We apply this approach to data collected in the mythic city of Angkor Thom (Cambodia), capital of the Khmer Empire between the 9th and 15th centuries. The first step in the process is to construct stable, interpretable pottery facies using a compromise-based clustering approach. In the second step, the predictions provided by supervised classification models make it possible to integrate less reliable pottery assemblages that are essential to the overall construction of the chronological model of the Angkor Thom city. Our approach offers the advantage of automatically processing large volumes of data and integrating uncertainty into the forecasts obtained for the least chronologically reliable sets.