CALRCN: a context-aware attentive long-term recurrent convolutional network for transportation mode identification based on GPS trajectory data

Transportation Mode Identification (TMI) based on GPS trajectory data is vital for data-driven urban mobility analysis and sustainable transportation planning. Despite recent advances in deep learning-based spatiotemporal modeling, effectively integrating heterogeneous contextual information into TMI remains a challenge. This paper proposes CALRCN, a Context-aware Attentive Long-term Recurrent Convolutional Network that jointly models spatial, temporal, and contextual characteristics of human mobility. The framework employs a structured input representation that separates GPS-based motion features from GIS-based contextual features. Spatial patterns are extracted using a convolutional module with channel-wise attention, while long-term temporal dependencies are captured through a Bidirectional Long Short-Term Memory (BiLSTM) with additive attention. Weather information is further incorporated to enhance contextual awareness. Experiments on the Geolife dataset demonstrate that CALRCN achieves an accuracy of 95.7%, outperforming existing deep learning baselines. Further evaluation supports the framework’s generalizability, while ablation studies validate the effectiveness of context fusion and attention mechanisms.