Optimizing resource-constrained urban management: a LightGBM approach for high-fidelity prediction

Urban analytics has emerged as a transformative tool for addressing the challenges of urbanization and fostering sustainable smart cities. Despite its growing significance, a gap remains in the literature on integrating advanced machine learning models, such as LightGBM, into real-world urban applications, particularly in resource-constrained environments. This paper proposes a unified framework that combines three urban analytics tasks: air quality forecasting (regression), traffic congestion nowcasting (classification), and energy demand nowcasting, using a consistent LightGBM methodology. The novelty lies in its multi-task formulation, which leverages shared preprocessing and feature sets. The framework also integrates heterogeneous data sources, including IoT sensors, social media, geospatial, and demographic data, while emphasizing computational efficiency and interpretability for deployment in limited-infrastructure settings. On real-world datasets, the PM2.5 forecasting model achieved an RMSE of 3.25 μg/m3 and an R2 of 0.88, while traffic congestion nowcasting attained a macro F1-score of 0.92. The energy demand nowcasting task identified peak usage patterns. These results demonstrate the potential of machine learning to support urban sustainability. However, challenges related to data privacy, algorithmic bias, and computational constraints must be addressed. The proposed framework provides actionable guidance for policymakers, city planners, and researchers, supporting the development of inclusive, resilient, and sustainable urban environments.