Complex Statistical Indicators Based On Symbolic Data Analysis: a straightforward and explainable approach to AI-supported Official Statistics
Conference
66th ISI World Statistics Congress
Format: WSC 2027: IPS Abstract
Keywords: symbolic_data_analysis
Abstract
The growing availability of heterogeneous data from multiple sources calls for new approaches to the construction of composite and multidimensional indicators in official statistics. This paper presents Symbolic Data Analysis as a framework for representing and synthesising complex information through variables with distribution values, interval values and set values. By preserving the variability and uncertainty of the data, symbolic representations can support the development of richer and more informative indicators.
The paper proposes indicators that evolve and combine elementary indicators across time to highlight inequalities between geographical areas, with a particular focus on Italy. The analysis draws on data from ISTAT, ISPRA and Copernicus data sources. A key innovative aspect is the definition of context-dependent weights, designed to reflect the specific territorial and temporal conditions in which the indicators are constructed and interpreted. This approach aims to make comparisons more meaningful, whilst preserving the complexity of the underlying information.
Particular attention is also paid to how symbolic representations can support explainable AI tools for the construction of indicators, as well as for policy monitoring and evaluation. By combining transparent weighting choices with representations that preserve variability and uncertainty, the proposed framework can contribute to more interpretable, transparent and reliable statistical production processes.