A Data-Driven Simulation and Machine Learning Framework for Shopping Trip Forecasting with Spatial Clustering

Retailing plays a pivotal role in the functioning of urban systems. While upstream supply chain activities such as manufacturing and distribution primarily affect freight movement, the retail interface translates consumer demand into individual travel behavior, shaping local traffic conditions and feeding back into upstream logistics. Despite its importance, shopping-related travel remains under-modeled in urban mobility research. To address this gap, this study develops a purpose-specific travel forecasting and simulation framework for predicting shopping trip demand in urban areas. The forecasting model integrates commercial-environment attributes, trip characteristics, and sociodemographic factors. A suite of machine learning (ML) models is evaluated, and the best-performing model is selected for the proposed simulation. Microlevel predictions are then scaled to the full urban region, followed by zonal aggregation and k -means spatial clustering to identify distinct retail-demand patterns and support scenario testing. Numerical results show that the random forest model outperforms alternative ML classifiers and, when implemented in the simulation, generates a citywide estimate indicating that shopping trips represent 14.3% of all weekday travel, in line with external regional benchmarks. The combined ML–simulation framework demonstrates strong predictive performance and reveals meaningful spatial and behavioral insights relevant to policymaking and planning applications. Although applied to Halifax, the modular structure of the framework makes it transferable to other urban regions and adaptable to additional trip purposes, supporting future extensions involving multiactivity modeling, causal impact analysis, and integration with passive mobility datasets.