A machine learning approach for the analysis of demographic features of the socio-economic deprivation at sub-municipal level

Abstract

This study provides a machine learning analysis of the social and demographic features of territories, associated with different levels of deprivation, measured by the Socio-Economic Deprivation Index (SED-I). The SED-I has been developed by the Italian National Institute of Statistics (Istat) and is aimed at measuring socio-economic and educational deprivation at sub-municipal level. In this paper, we exploit the availability of socio-economic indicators and demographic context variables at Sub-Municipal Areas (SMAs) level to provide useful information for policies aimed at contrasting the economic, social, and educational deprivation as measured by the SED-I. In the context of socio-economic deprivation, territories may be characterised by a variety of interrelated variables: indicators related to the level of education in the area, employment, household structure, demographic composition referring to population density, foreign subpopulations, and age, and finally, social conditions, are employed in the analysis. Principal Component Analysis (PCA) is employed to summarise the information contained in this large number of correlated indicators. This unsupervised learning method allows us to analyse the relationships among these variables and the SED-I, pointing out the differences between two different socio-economic environments such as Trieste and Cagliari. Furthermore, a K-means cluster analysis is employed to identify population groups with similar composition in relation to deprivation and demographic features. In the K-means cluster analysis, Elbow and Silhouette methods are used to choose the optimal number of clusters. The analysis is carried out focusing on Trieste and Cagliari data, to provide a robust application on different socio-economic contests.