MACHINE LEARNING FOR URBAN ENVIRONMENTAL DAMAGE VALUATION IN DATA-SPARSE CITIES: A SYSTEMATIC LITERATURE REVIEW AND CONCEPTUAL FRAMEWORK FOR SAMARKAND, UZBEKISTAN

Abstract

Background: Ambient fine particulate matter (PM2.5) pollution in Uzbekistan cities has reached crisis proportions, yet the economic burden it imposes remains severely underestimated by the traditional econometric tools on which policymakers depend. Uzbekistan’s national annual PM2.5 average of 41.2 μg m⁻³ — more than eight times the World Health Organization guideline of 5 μg m⁻³ — signals an acute public health emergency that demands more precise quantification methods. Samarkand, Uzbekistan’s second largest city, embodies these challenges yet has received virtually no dedicated air quality modelling attention in the peer-reviewed literature. Objectives: This systematic literature review pursues three objectives: (1) to synthesise evidence on how machine learning (ML) algorithms — particularly Random Forest (RF) and Extreme Gradient Boosting (XGBoost) — compare with conventional econometric models in urban PM2.5 prediction; (2) to examine how ML-derived exposure estimates propagate into economic damage metrics, specifically Disability-Adjusted Life Years (DALY), Quality-Adjusted Life Years (QALY), and Value of Statistical Life (VSL); and (3) to propose a conceptual research framework tailored to Samarkand’s data environment. Methods: Following PRISMA 2020 guidelines (Page et al., 2021), we searched Scopus, Web of Science, PubMed, and Google Scholar for studies published between January 2019 and January 2025. Boolean operators combined three thematic blocks: ML methodology, urban air pollution, and economic or health valuation. An initial yield of 4,847 records was progressively reduced through duplicate removal, title/abstract screening, and full-text eligibility assessment to a final corpus of 63 peer-reviewed studies.