Clustered embedding using deep learning to analyze urban mobility based on complex transportation data.

Area of Science:

• Urban planning and transportation science.
• Data science and artificial intelligence.

Background:

• Urban mobility significantly shapes city development and influences economic and social progress.
• Analyzing urban mobility patterns aids in predicting traffic flow and public transport usage.
• Deep learning embeddings show promise for urban mobility pattern analysis due to their ability to handle complex, large-scale data.

Purpose of the Study:

• To address the limitations of single-vector embeddings in capturing complex urban mobility patterns.
• To propose a novel deep learning method for analyzing urban mobility by clustering and embedding patterns.
• To enhance the understanding of both personalized and collective urban mobility.

Main Methods:

• Clustering urban mobility patterns based on spatiotemporal characteristics.
• Embedding clustered mobility patterns to capture implicit meanings.
• Utilizing deep learning for pattern extraction and analysis of large-scale transportation data.

Main Results:

• The proposed method achieved top-1, 3, and 5 accuracies of 73.64%, 88.65%, and 91.54% in predicting successive points of interest (POIs).
• These results significantly outperform conventional methods, which achieved 59.48%, 75.85%, and 80.1% respectively.
• The method enables effective analysis of urban mobility through arithmetic operations on POI vectors.

Conclusions:

• The novel deep learning approach effectively analyzes complex urban mobility patterns.
• Clustering and embedding mobility patterns provide deeper insights into individual and collective behavior.
• The method offers a significant advancement in predicting points of interest and understanding urban dynamics.