XAI and geographic information: application to paleoenvironmental reconstructions

Bastien Zimmermann, Matthieu Boussard

Abstract:

This work shows the contribution of explicability to the reconstruction of paleoenvironments. Integrating explanations to a predictive approach provides the domain expert additional information and trust. It aims at building a model allowing, from excavation data, to infer the environment corresponding to a given layer and a given paleolithic period. In this context, the prediction of the model alone has less value than the underlying explanations that allow archaeologists to question their assumptions. Due to the uncertainties in the data, this work focuses more on data-oriented explanations tools, such as data-Shapley. Finally, a contribution of this article is the use of a Geographic Information System, allowing us to exploit to the maximum the information we can obtain from the explainability tools.