A semi-supervised multi-criteria sorting approach to constructing social vulnerability composite indicators

IPVS (São Paulo Social Vulnerability index) was created by the State government of São Paulo, Brazil, with the identification and spatial location of the areas that contain the population segments most vulnerable to poverty. IPVS relies on a data-driven approach which is implemented by means of multivariate analysis techniques such as principal component analysis. A limitation of such a statistical approach is that it only considers information brought by data, as it does not take into consideration subjective information provided by decision makers. Motivated by this limitation, we propose an alternative approach based on multi-criteria sorting. For this purpose, we introduce a conceptual sorting framework based on the SMAA methodology and on the Choquet integral, which allows us to take into consideration interactions between criteria. The proposed sorting scheme classifies the municipality regions into groups characterized by reference values previously defined by the decision maker. As an important result, we show that our proposal provides more flexibility for vulnerability analysis in the sense that it allows decision makers to delve into different scenarios, opening the way for customized decision strategies.