An Ensemble Learning-Based Data-Centric Geospatial Framework of Property Value Index Estimation Based on Satellite-Derived Features

Abstract

The proper estimation of property value is an intricate issue that is affected by socioeconomic, environmental, and infrastructural variables. Conventional methods of valuation are based mainly on formal housing information, including income, population, and property type, and usually do not reflect the spatial and environmental context that has a great influence on the desirability of real estate.
This paper suggests a data-driven model to make a Property Value Index (PVI) estimate based on incorporating structured housing data from the California Housing dataset and satellite-based geospatial attributes. This study is in contrast with the traditional methods of optimization of the models, it is more about the systematic construction of a spatially enriched dataset based on the remote sensing data sources. The most important attributes, such as vegetation density (NDVI), light intensity at night, road density, and the presence of water, are derived using the geospatial computing and aggregated across the localized spatial buffers using Google Earth Engine.
Ensemble learning models such as a baseline Random Forest, an enhanced Random Forest and XGBoost are used to evaluate the proposed dataset. The experimental findings indicate that the use of geospatial features makes a significant contribution to the predictive performance, and XGBoost can lead to the greatest results (R2 =0.80), which is better than the traditional methods.
Moreover, the spatial validation establishes that the predicted values follow the geographical trends in the real world; that is, regions with an increased economic activity and density in infrastructure have a high PVI score.
The results point to the importance of feature engineering and data representation rather than the complexity of the models in the context of property valuation. This work offers an extensible and practical system to real estate analytics and location intelligence systems and highlights the significance of incorporating geospatial intelligence in predictive modelling.