Effective trajectory selection and classification are pivotal in user tracking systems utilizing spatiotemporal data collected from city sensors. However, the inherent limitations in sensor technologies and data collection point distributions often result in low-quality spatiotemporal data. Real-life trajectory classification encounters challenges due to the following: (1) high-order and sparse activity data encompassing both temporal and spatial contexts, and (2) inherent vagueness in the semantics of visited locations, making it difficult to represent behavioral intentions. Traditional statistics-based or trajectory-based feature approaches prove ineffective with non-discriminate features. In response to these challenges, we introduce a novel classification method that integrates fuzzy spatiotemporal features and crowd habit features. This approach involves feature extraction using the Time-Geo Hash (TGH) and User Transit Pattern and Similarity (UTPS) models, followed by the training of a machine learning classification model. On the basis of the performance indicators of classification models, we identify two classification algorithms, incorporate the Bagging algorithm from ensemble learning to enhance the UTPS classification model, and combine the TGH and UTPS models through specified rules. Extensive experiments demonstrate that our proposed model significantly outperforms other classification baselines when applied to a labeled real-life dataset, emphasizing its effectiveness in handling noisy and challenging spatiotemporal data for trajectory classification in user tracking systems.