Fuzzy clustering of spatial interval-valued data

Pierpaolo D’Urso,Livia De Giovanni,Lorenzo Federico,Vincenzina Vitale

DOI: https://doi.org/10.1016/j.spasta.2023.100764

IF: 2.125

2023-10-01

Spatial Statistics

Abstract:In this paper, two fuzzy clustering methods for spatial interval-valued data are proposed, i.e. the fuzzy C -Medoids clustering of spatial interval-valued data with and without entropy regularization. Both methods are based on the Partitioning Around Medoids (PAM) algorithm, inheriting the great advantage of obtaining non-fictitious representative units for each cluster. In both methods, the units are endowed with a relation of contiguity, represented by a symmetric binary matrix. This can be intended both as contiguity in a physical space and as a more abstract notion of contiguity. The performances of the methods are proved by simulation, testing the methods with different contiguity matrices associated to natural clusters of units. In order to show the effectiveness of the methods in empirical studies, three applications are presented: the clustering of municipalities based on interval-valued pollutants levels, the clustering of European fact-checkers based on interval-valued data on the average number of impressions received by their tweets and the clustering of the residential zones of the city of Rome based on the interval of price values.

remote sensing,geosciences, multidisciplinary,mathematics, interdisciplinary applications,statistics & probability