Aprendizagem de máquina e modelagem estatística na avaliação de imóveis urbanos: a influência de variáveis ambientais em Palmas-TO

Abstract

This research analyzed the influence of environmental variables on urban property prices in the city of Palmas-TO, with emphasis on the proximity to Cesamar Park and the reservoir of the Lajeado Hydroelectric Plant, through the application of traditional statistical models and Machine Learning (ML) techniques. The central motivation lies in the search for models capable of more accurately capturing the effect of environmental amenities on the real estate market, contributing to the incorporation of environmental value into economic analysis and, eventually, into public policies, such as the collection of property tax (IPTU). The investigation also falls within the field of urban planning, by recognizing that real estate appreciation is directly linked to location, accessibility, and the availability of urban services and infrastructure, in addition to the presence of environmental assets. Thus, understanding how these elements influence property prices allows for better-informed territorial planning strategies, supporting public policies aimed at socio-spatial equity and the sustainability of Palmas’ urban expansion. Linear Models (LM) and Generalized Linear Models (GLM), as well as fourteen ML algorithms, were estimated with the purpose of comparing the predictive performance of each approach. The explanatory variables included physical, locational, socioeconomic, and environmental attributes, the latter represented by continuous distances and proximity dummies to environmental assets. The results indicate that ML models outperformed traditional models in terms of predictive capacity and sensitivity in identifying the effects of environmental variables on property prices. These models demonstrated greater robustness and flexibility in capturing complex and nonlinear relationships, expanding analytical possibilities in the field of real estate valuation. Among the tested algorithms, Bagged Trees, Random Forest, and Boosted Trees stood out as particularly effective in modeling nonlinear relationships and identifying complex interactions among attributes. The findings reinforce that the application of Machine Learning techniques constitutes a consistent and innovative methodological approach to urban property valuation, especially when seeking to incorporate environmental variables into the price formation dynamics of the real estate market, overcoming the limitations of traditional models.