Dynamic parking demand patterns modeling with PSO-XGBoost: Implications for parking management

Understanding the characteristics of on-street parking demand and its generation mechanisms is crucial for parking planning and management. However, existing studies have largely overlooked the temporal dynamics of on-street parking demand and its relationship with the built environment. This gap limits to comprehend the spatial mechanisms driving on-street parking demand, leading to an insufficient theoretical foundation for parking practice. To address this issue, first, we collected over 800,000 parking records from the intelligent on-street parking system and proposed a preprocessing method to extract parking demand features from the original data. Subsequently, we constructed a Dynamic Time Warping (DTW) model to cluster the temporal patterns of on-street parking demand. Finally, to mitigate the imbalance between the number of samples and variables, we proposed an XGBoost model optimized using Particle Swarm Optimization (PSO). This model establishes the relationship between the built environment and parking demand patterns. The results of model were further interpreted through relative feature importance and partial dependence plots. The proposed approach and findings provide valuable insights for the planning and management of on-street parking.