Explainable spatial machine learning for hedonic real estate modeling

Accurately modeling rents and prices is a key challenge in real estate analysis. Traditional linear models may fail to capture complex non-linear relationships, and spatial dependencies are often ignored in existing machine-learning approaches. This article introduces a novel hybrid statistical machine-learning model for modeling real estate rents and prices. The proposed approach combines a spatial Gaussian process with tree boosting. In so doing, spatial correlations are explicitly accounted for, and the tree-boosting part can handle complex non-linear relationships and interactions. We compare the proposed model against established benchmarks using a large-scale dataset consisting of more than 1.5 million rental apartment listings across Germany and also a smaller condominium price listings dataset. Our findings demonstrate that the proposed model yields superior prediction accuracy due to accounting for both nonlinear patterns and spatial dependencies. We further use machine-learning explainability techniques to better understand the nonlinear relationships present among rents and predictor variables, and we conduct a detailed analysis of the impact of locational characteristics on rents.