Dimensionality reduction for data visualization and exploratory analysis of ceramic assemblages

Dimensionality reduction algorithms are essential in the study of multivariate datasets. Many variables make it difficult to visualize data. In Archaeology, this problem often concerns the study of some variables, which can be quantitative or qualitative. In this article, several methods for dimension reduction are applied to a pottery dataset from the protohistoric necropolis Osteria dell’Osa, located 20 km East of Rome. These methods offer the possibility of visualising and analysing large amount of data in a very short time. Our results show that non-linear and non-parametric algorithms such as t-SNE and UMAP are the best choice for visualising and exploring this type of data.