Maximum likelihood estimation of spatially varying coefficient models for large data with an application to real estate price prediction

In regression models for spatial data, it is often assumed that the marginal effects of covariates on response are constant over space. practice, this assumption might be questionable. article, we show how a Gaussian process-based spatially varying coefficient (SVC) model can estimated using maximum likelihood estimation (MLE). addition, present an approach scales to large data by applying covariance tapering. We compare our methodology existing methods such as Bayesian stochastic partial differential equation (SPDE) link, geographically weighted (GWR), and eigenvector filtering (ESF) in both simulation study application where goal predict prices real estate apartments Switzerland. The results from MLE increased predictive accuracy more precise estimates. Since use model-based approach, also provide variances. contrast approaches, method better number points moderately-sized, e.g., above ten.