A. Evangelista, C. J. Acal González, A. M. Aguilera del Pino, A. Sarra, T. Di Battista, S. Palermi
Variable selection is one of the hardest issues in multivariate regression models. Several methods have been proposed to select smaller subsets of informative variables. In this work, a new variable selection method is proposed in the context of multiple function-on-function linear regression (MFFLR) model. By merging functional data analysis with group Lasso method, a principal components’ regression approach has been adopted to estimate the MFFLR model.
The proposed technique allows to reduce a large number of discrete observations highly correlated for each curve to a functional form that conserves all relevant information and obtain a more sensible and interpretable model for the environmental problem at hand.
The method has been tested with real data obtained from the automatic reporting platform run by Regional Agency for the Environmental Protection of Abruzzo Region (Italy) with the aim of investigating the relationship of PM10 with other interesting meteorological variables.
Keywords: variable selection; group Lasso; functional data analysis; environmental data.
Scheduled
Invited session Spanish-Italian
June 9, 2022 10:10 AM
Conference hall