A hierarchical temporal attention-based LSTM encoder-decoder model for individual mobility prediction.

Area of Science:

• Artificial Intelligence
• Data Science
• Human Mobility Analysis

Background:

• Individual mobility prediction is vital for applications like traffic planning and location advertising.
• Current methods primarily focus on short-term predictions and lack the ability to capture long-term dependencies.
• Understanding daily and weekly mobility patterns is essential for accurate long-term forecasting.

Purpose of the Study:

• To propose a novel hierarchical temporal attention-based LSTM encoder-decoder model for individual location sequence prediction.
• To effectively capture both long-term and short-term dependencies in individual mobility trajectories.
• To uncover and interpret periodical mobility patterns by integrating calendar cycles.

Main Methods:

• Developed a hierarchical attention mechanism combining local (daily) and global (weekly) temporal attention.
• Utilized a Long Short-Term Memory (LSTM) encoder-decoder architecture.
• Incorporated calendar cycles to model individual travel regularities.

Main Results:

• The proposed model significantly outperforms four baseline methods across three evaluation metrics.
• Demonstrated superior performance on individual trajectory datasets with varying degrees of traveling uncertainty.
• The hierarchical attention mechanism successfully captures both short-term and long-term dependencies.

Conclusions:

• The novel hierarchical temporal attention model provides accurate long-term individual location sequence prediction.
• The model offers interpretable insights into daily and weekly mobility patterns through attention weight visualization.
• This approach enhances human mobility applications by understanding complex travel regularities.