A0734
Title: Matrix completion for spatiotemporal air quality datasets
Authors: Rodolfo Metulini - University of Bergamo (Italy) [presenting]
Abstract: Matrix completion (MC) is a recent and flexible statistical learning method for imputing missing values and performing counterfactual analysis in structured datasets. A key advantage of MC methods is that they avoid relying on stringent model assumptions. A widely used approach leverages nuclear norm regularization, which promotes low-rank approximations of the data matrix. Recent advances on nuclear norm MC incorporate unit- and time-specific fixed effects, which is crucial to avoid biased imputations in panel data where strong heterogeneity across units and time is expected. The performance of existing MC methods is evaluated on panel data from the ARPA Lombardia air quality monitoring network. Given the spatial correlation inherent in air pollution data, the aim is also to incorporate spatial constraints into the matrix completion framework and to evaluate model accuracy and interpretability. The development and testing of MC methods contribute to the broader goal of conducting accurate counterfactual analyses in policy evaluation, particularly for assessing the impact of mobility-related interventions on air pollution levels. By enabling reliable estimations of air quality in the absence of specific policies, such methods can support informed decision-making in environmental and urban planning.
