CSGAN: Modality-Aware Trajectory Generation via Clustering-based Sequence GAN.

Area of Science:

• Urban Planning
• Transportation Science
• Data Science

Background:

• Human mobility data is vital for urban planning, transportation, and public health.
• Privacy and data quality issues hinder the collection and sharing of real-world trajectories.
• Generative Adversarial Networks (GANs) are increasingly used to create synthetic trajectories.

Purpose of the Study:

• To improve synthetic trajectory generation by explicitly modeling transportation modalities.
• To generate more diverse, representative, and realistic trajectories.
• To preserve geographical density, trajectory, and transition properties in synthetic data.

Main Methods:

• Proposed a Clustering-based Sequence Generative Adversarial Network (CSGAN).
• CSGAN simultaneously clusters trajectories by modality and learns real-world trajectory properties.
• Developed new metrics for evaluating trajectory effectiveness, including modality distribution and transition probabilities.

Main Results:

• CSGAN generates more diverse and representative trajectories across different transportation modes.
• The model preserves key properties like geographical density and transition levels.
• Experimental results demonstrate CSGAN's superiority over state-of-the-art models on real-world datasets.

Conclusions:

• Explicitly capturing transportation modalities enhances synthetic trajectory generation.
• CSGAN offers a robust method for creating realistic and representative human mobility data.
• The proposed evaluation metrics provide a comprehensive assessment of synthetic trajectory quality.