Answers are grounded in this paper and cite their sources.
Efficient and accurate Transportation Mode Identification (TMI) is essential for modern urban planning and traffic management, enabling optimized mobility and promoting transportation sustainability. While many existing TMI methods employ deep learning, they often rely on manual feature engineering, which may cause information loss and limit the exploitation of rich trajectory patterns. This study introduces a parameter-efficient Convolutional Neural Network (CNN) that leverages a novel spatiotemporal encoding of GPS trajectories. By transforming the data into structured three-channel sequences—capturing changes in location and time (ΔX, ΔY, ΔT)—the proposed method removes the need for handcrafted features and enables the network to autonomously extract high-level representations, while minimizing the risk of information loss. Additionally, temporal (e.g., holidays and time-of-day) and spatial (e.g., urban districts) attributes are incorporated to enrich the model’s input and enhance its ability to recognize complex mobility behaviors. Experimental results show that the proposed ensemble model achieves an accuracy of 89.5%, surpassing prior related methods. Importantly, the number of trainable parameters has been significantly reduced—from 2.6 million to approximately 1 million—without compromising performance. This research underscores the importance of spatiotemporal context in trajectory data, questions the continued reliance on hand-crafted features, and shows that prior models may have introduced unnecessary complexity—demonstrating that a lightweight CNN can achieve superior performance with far fewer parameters.
No takes yet. Share an insight, caveat, or question.
tayebi et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: