Use of Urban Lighting Data for Real Estate Price Forecasting

This paper presents an improved model for forecasting rental property values in large metropolitan areas by incorporating features derived from the spatial distribution of urban lighting. The study uses rental housing data from Houston and Los Angeles (USA) together with high-resolution nighttime satellite imagery to generate additional explanatory variables. Light clusters are identified from satellite images and processed to determine their geographic location and spatial relationships. The clusters are georeferenced by using the Quantum Geographic Information System, thus enabling integration with other spatial datasets and improving modelling accuracy. The paper describes the methodology for feature extraction, spatial clustering, and integration into machine learning workflows. A Light Gradient-Boosting Machine predictive model is developed and compared with baseline models. The experimental results show that the proposed approach reduces the mean squared error by 11.8% for Houston and 9.37% for Los Angeles relative to conventional models. The findings demonstrate the usefulness of nighttime illumination features for capturing socio-economic and spatial patterns relevant to urban rental markets. The proposed methodology highlights the potential of combining geospatial data and machine learning techniques to improve automated valuation models and support urban analytics and smart city planning.