# RECORD 05c8066a4f1ebf0565508520c4002c90671161c4 # TITLE: Algorithm-Driven Hedonic Real Estate Pricing – An Explainable AI Approach # YEAR: 2024.0 VENUE: Real Estate Management and Valuation DOI: 10.2478/remav-2025-0003 # AUTHORS: S. Tarasov; Bartłomiej Dessoulavy-Śliwiński # PAGES: 13 CANDIDATE TABLE PAGES (1-based): [] # ABSTRACT: Abstract Data-driven machine learning algorithms triggered a fundamental change in hedonic real estate pricing. However, their adaptive nonparametric structure makes inference and out-ofsample prediction challenging. This study introduces an explainable approach to interpreting machine learning predictions, which has not been done before in the local market context. Specifically, Random Forest and Extreme Gradient Boosting models are developed for residential real estate price prediction in Warsaw in 2021 on 10,827 property transactions. Model-agnostic Explainable Artificial Intelligence (XAI) methods are then used to investigate the black box decision making. The results show the practicability of applying XAI frameworks in the real estate market context to decode the rationale behind data-driven algorithms. Information about the relationships between input variables is extracted in greater detail. Accurate, reliable and transparent real estate valuation support tools can offer substantial advantages to participants in the real estate market, including banks, insurers, pension and sovereign wealth funds, as well public authorities and private individuals. ===== PAGE 1 (score 5, numeric-paren 0, table-words 0, var-words 5) ===== REAL ESTATE MANAGEMENT AND VALUATION - vol. 33, no. 1, 2025 eISSN: 2300-5289 | © 2024 The Author(s) | Article under the CC BY 4.0 license 22 ALGORITHM-DRIVEN HEDONIC REAL ESTATE PRICING – AN EXPLAINABLE AI APPROACH Stanislav Tarasov1, Bartłomiej Dessoulavy-Śliwiński2* 1 Faculty of Economic Sciences, University of Warsaw, Długa 44/50, 00-241 Warsaw, Poland, e-mail: swtarasow@gmail.com, ORCID: 0000-0001-7018-9847 2 Department of Management and Information Technology, Faculty of Economic Sciences, University of Warsaw, Długa 44/50, 00-241 Warsaw, Poland, e-mail: bsliwinski@wne.uw.edu.pl, ORCID: 0000-0002-4395-9815 * Corresponding author ARTICLE INFO ABSTRACT Keywords: Data-driven machine learning algorithms triggered a fundamental change in hedonic real estate pricing. However, their adaptive nonparametric structure makes inference and out-of- sample prediction challenging. This study introduces an explainable approach to interpreting machine learning predictions, which has not been done before in the local market context. Specifically, Random Forest and Extreme Gradient Boosting models are developed for residential real estate price prediction in Warsaw in 2021 on 10,827 property transactions. Model-agnostic Explainable Artificial Intelligence (XAI) methods are then used to investigate the black box decision making. The results show the practicability of applying XAI frameworks in the real estate market context to decode the rationale behind data-driven algorithms. Information about the relationships between input variables is extracted in greater detail. Accurate, reliable and transparent real estate valuation support tools can offer substantial advantages to participants in the real estate market, including banks, insurers, pension and sovereign wealth funds, as well public authorities and private individuals. residential real estate, machine learning, explainable artificial intelligence, mass appraisal, automated valuation models JEL Classification: C14, C18, R39 Citation: Tarasov, S. & Dessoulavy-Śliwiński, B. (2025). Algorithm-driven hedonic real estate pricing - An explainable AI approach. Real Estate Management and Valuation, 33(1), 22-34. https://doi.org/10.2478/remav-2025-0003 1. Introduction Real estate is the world’s biggest asset class, and residential real estate is a major asset for many individuals. Hence, property price prediction holds significant importance and increasingly attracts attention of researchers due to a variety of factors, including investment decisions, buy and sell transactions, tax considerations, or real estate investment trusts’ activity (Arribas et al., 2016). Due to market and regulatory changes in recent years, real estate is appraised more frequently than ever before, and inaccurate valuation can lead to various complications affecting property owners, buyers, and financing providers. Hence, there is a strong demand for accurate, speedy, and dependable support tools for real estate valuation. As a result, automated valuation model (AVM) development has been steadily on the rise in the industry and in academic research. On the one hand, the subject of predicting real estate prices is well recognized in the literature. Studies on this topic in the field of applied econometrics often employ conventional linear regression methods along with their spatial extensions, as well as ML-based models. Research has indicated varying results when comparing the performance of ML and spatial models, with some studies demonstrating superior results for the former (e.g., Rico-Juan & Taltavull de La Paz, 2021), and others presenting contrasting findings, in favor of the latter (e.g., Przekop, 2022). Li (2022), in their experimental study of extracting spatial effects from an ML model using a local interpretation method, shows that “locally interpreted machine learning models are good alternatives to spatial statistical models and perform better when complex spatial and non-spatial effects (e.g. non-linearities, interactions) coexist and are unknown”. On the other hand, following the reasoning of Angrist and Pischke (2010), “empirical evidence on any given causal effect