Non-linearities and interaction effects in Belgian residential rents: an explainable artificial intelligence approach

Grounded in hedonic pricing theory, this study examines non-linear and heterogeneous relationships in residential rents. Using 78,788 Belgian rental listings from 2022 as an empirical application, we compare a linear Ridge model with CatBoost and interpret CatBoost using SHapley Additive exPlanations (SHAP). CatBoost predicts rents more accurately, while SHAP dependence plots suggest a threshold-like association with energy consumption, a U-shaped build-year profile and housing-type heterogeneity in the mobility-score relationship. These findings illustrate how flexible, black-box models with SHAP can complement conventional hedonic specifications.