Development of a methodology for estimating average and median market prices of residential real estate and rents by local government units using machine learning methods

Commissioned by: Ministry of Physical Planning, Construction and State Assets
Project duration: December 4, 2025 - January 17, 2026
Project manager: Maruška Vizek


Summary: 
This project develops a comprehensive machine learning  methodology for estimating average and median market prices of residential real estate and rents at the level of local government units using machine learning methods. The core objective is to address the problem of missing or unreliable price data in areas with low transaction volumes by combining spatial information, socio-economic indicators, housing stock characteristics, infrastructure accessibility, and market activity variables.

The methodology integrates spatial features, including spatial lags based on neighbouring local government units using KNN method, and applies robust imputation techniques to ensure full territorial coverage. Multiple predictive machine learning models are estimated and systematically compared, including linear regression, regularized regressions (LASSO, Ridge, Elastic Net), regression trees, random forests, and gradient boosting models. Model performance is evaluated using out-of-sample validation and standard accuracy metrics, while model-agnostic and algorithm-specific methods are used to assess variable importance and identify key price determinants.

In addition, parsimonious core models and hierarchical mixed-effects models are developed to improve interpretability and ensure stable predictions across heterogeneous local goverment units. The final output consists of  replicable, consistent and comparable estimates of residential property prices and rents for all local government units, providing a transparent, data-driven basis for housing market monitoring, policy design, and evidence-based decision-making.