Prediction Model Establishment for Residential Community Occupancy Considering Urban Environment

The accurate prediction of occupancy rates within the residential sector is pivotal for urban and energy planning and management. This study aims at addressing the often-neglected influence of regional variations within urban settings on occupancy rates by employing multi-source urban data and explainable machine learning algorithms. Utilizing mobile signaling data from January and July 2022 in Guangzhou, we developed prediction models using the CatBoost algorithm with hyperparameters finely tuned via Bayesian optimization. The SHapley Additive exPlanation was applied to elucidate the influence of different urban configurations. Our findings reveal significant seasonal variations and distinct differences between weekdays and weekends, showing higher occupancy rates in winter than in summer and elevated rates during weekends. Additionally, significant fluctuations were observed among different age groups on weekdays. The prediction models exhibited robust capabilities, achieving R2 values exceeding 0.95 for weekdays and 0.91 for weekends. Key findings highlight that the density of points of interest related to shopping and food, along with variability in building heights, significantly impact occupancy rates. The influence of diverse urban configurations varies across different hours of the day. By the comprehensive analysis of regional variations and the application of advanced machine learning techniques to predict and explain residential occupancy rates, this research offers tools to optimize the management and sustainability of built environments, enhances the understanding of urban operation patterns and contributes to advancing smart city technologies, providing valuable insights for urban and energy planning.