Organizer: Luzi
Chair: Luzi
Arosio, F. M., Di Zio, M., Massacci, G., Ortame, F., Pappagallo, A., Pugliese, F., Salamone, N., Sisti, F., Talice, S., Valentino L. Using Automatic Identification System data for official statistics
Abstract
The Automatic Identification System (AIS) is a ship tracking technology designed to enhance safety and navigation management. It provides positions of vessels at regular and frequent interval times. National statistical institutes are investing in the study of those data with the aim of improving their maritime transport statistics. In this paper, the analysis and the algorithms used for their transformation for official statistical purposes in the Istat context are illustrated and discussed.
De Cubellis, M., Foschi, F., Mercorio, F., Mezzanzanica, M., Nobani, N., Pugliese, F. Explainable AI in an official statistics domain
Abstract
Machine Learning (ML) models are increasingly used in Official Statistics (OS) for their ability to discover complex patterns in data. However, the adoption of ML in this domain requires high standards of transparency, reproducibility, and compliance with ethical norms, underscoring the need for eXplainable Artificial Intelligence (XAI). This paper investigates the application of SHAP (SHapley Additive exPlanations) to interpret classification models applied to the “Trips and Holidays” survey dataset from ISTAT. While the performance metrics of the considered four models were statistically similar, SHAP analysis revealed significant variations in feature contributions to the decision-making. This highlights the importance of understanding model behaviour beyond accuracy metrics to ensure informed and reliable applications in official statistics.
Di Consiglio, L., Pichiorri, T., Piovani, A., Tuoto, T., Zhang, L. Mobile Network Operator data meet official statistics: a use case on commuting statistics
Abstract
Mobile network operator (MNO) data have great potentials for producing official statistics on population, tourism, mobility and environment. However, MNO data would not suffice on their own whenever the target statistical unit or the measurement unit is not mobile device per se. In this work we explore the use of Mobile Network Operator (MNO) data integrated with surveys to produce commuting statistics, traditionally estimated with census data. Availability of MNO data allows for the derivation of recurrent flows between home and work/study locations. These flows serve as a valuable information and can be used as the primary source for producing official statistics by adjusting MNO data coverage with a quasi randomisation approach. On the other hand superpopulation model can be exploited to describe the phenomenon. These methods are applied to estimate the number of commuters of each municipalities of an Italian region.