Conformal prediction of real estate prices with machine learning

Uncertainty quantification associated with real estate appraisal has largely been ignored in the literature. The aim of this dissertation is to fill this gap by analysing uncertainty in property valuation using machine learning complemented by conformal prediction. Conformal prediction quantifies uncertainty associated with individual predictions and provides a range of possible outcomes around point estimates based on a pre-defined significance level. By applying conformal quantile regression, we can mitigate limitations of early conformal regression approaches and we are able to build intervals that only require the data to be exchangeable for the coverage to be guaranteed. Through an empirical study of property prices in the San Francisco Bay Area, we find that the conformal quantile regression provides adaptive prediction intervals with guaranteed coverage that captures uncertainty variations across different property prices.