A0944
Title: SPDE-Forest: An hybrid approach for modeling geostatistical data
Authors: Michela Cameletti - Universita degli Studi di Bergamo (Italy)
Luca Patelli - University of Pavia (Italy) [presenting]
Mario Figueira Pereira - Universitat de Valencia (Spain)
Abstract: The aim is to propose SPDE-Forest, a hybrid two-stage approach for modeling geostatistical data and performing spatial prediction. The proposed strategy combines the random forest algorithm and the stochastic partial differential equations (SPDE) approach implemented through the INLA algorithm. SPDE-Forest is able to model complex non-linear relationships between the response variable and the predictors, thanks to the random forest stage, while accounting for the residual spatial correlation by means of the SPDE part. The out-of-sample predictive performance of SPDE-Forest is assessed through a simulation study and compared to that of existing competitors.
