Understanding carsharing user Behavior: A spatiotemporal approach to stop activity classification

Understanding how travelers use carsharing services is critical for optimizing operations and guiding urban mobility policy. Traditional data collection methods, such as surveys and trip diaries, are resource-intensive and prone to recall bias. This study presents a stop-level activity classification framework for round-trip carsharing in Bangkok, Thailand, integrating GPS trajectories, trip metadata, land use, and proximity to points of interest (POIs). Ground truth was obtained through structured phone interviews with 217 users within 48 h of trip completion. Three machine learning models—Support Vector Machine (SVM), Random Forest (RF), and Extreme Gradient Boosting (XGBoost)—were evaluated, with XGBoost achieving the highest overall accuracy of approximately 63%. The model performed particularly well in identifying activities with distinct spatial–temporal patterns, including Personal Errands and Escort Activities and Outdoor & Leisure with accuracies of around 70%. Feature importance analysis indicated that stop duration, trip extent beyond Greater Bangkok, and POI density were key predictors. These results can support operators and planners by providing insights into how users interact with different locations, helping to inform service planning, pricing strategies, and infrastructure allocation. By focusing on developing urban markets in Southeast Asia, the study contributes regionally grounded evidence to the broader literature on shared mobility, digital data collection, and sustainable transport.