Evaluation of Annotation Ambiguity in Common Supervised Machine Learning Classification Approaches for Cultural Heritage

Aim of the present work is to combine the semantic classification approaches proposed for architectural heritage, exploiting Machine Learning (ML), with tools that take into account the uncertainty associated with the operator’s annotation of the training data, in order to encode the ambiguity or fuzziness of the labeling procedure. We aim at defining the quality of the information associated with the manual anno-tation, accordingly. The experimental framework for this research was established within the Level II University Master Course UNIBIM—Master BIM Manager, at the University of Pisa, during the academic year 2023–24. Students were assigned a manual annotation exercise involving the same architectural object: the primary objective was to describe and discern variations in annotations among individuals with diverse cultural backgrounds and educational experiences.