Ensemble Prediction via Covariate-dependent Stacking
This study introduces Covariate-dependent Stacking (CDST), a novel ensemble prediction method where model weights flexibly change based on input covariates. CDST outperforms traditional stacking and averaging by optimizing these weights using cross-validation. It proves highly effective for complex spatio-temporal problems, such as land price prediction, especially when base models fail to capture underlying data complexity.