Can ensemble learning reveal real estate value – a study of England and Wales cases anonymous

This paper introduces a high-fidelity Automated Valuation Model (AVM) that resolves the trade-off between predictive accuracy and reliable uncertainty quantification. Applying AutoGluon, a multi-layered ensemble learning system, to a large-scale dataset of property transactions in England and Wales (2011–2019), our model achieves state-of-the-art accuracy with R2 values exceeding 0.96 for houses and 0.86 for flats, significantly outperforming standard benchmarks. Beyond single-point estimates, our framework generates robust prediction intervals for comprehensive risk assessment. We also use SHAP to ensure transparency by identifying key value drivers like property size and energy efficiency. Our findings establish that superior accuracy is the necessary foundation for generating both meaningful model interpretations and trustworthy uncertainty metrics, providing a robust tool for mortgage lending, investment analysis, and housing policy.