Human mobility is well described by closed-form gravity-like models learned automatically from data.

Area of Science:

• Urban studies
• Computational social science
• Data science

Background:

• Human mobility modeling is crucial for urban planning, sustainability, public health, and economic development.
• Current models like gravity models offer interpretability but limited accuracy, while complex machine learning models provide higher accuracy but lack behavioral insights.

Purpose of the Study:

• To develop and evaluate simple, interpretable machine-learned models for human mobility flow prediction.
• To demonstrate that these models can match or exceed the accuracy of complex machine learning approaches while offering better interpretability.

Main Methods:

• Development of closed-form, machine-learned mobility models.
• Comparison of model performance against traditional gravity models and complex machine learning models using diverse datasets.
• Evaluation of model interpretability and extrapolation capabilities.

Main Results:

• Simple machine-learned models achieve prediction accuracy comparable to complex machine learning models.
• These novel models demonstrate superior extrapolation capabilities compared to existing methods.
• The developed models are interpretable, similar to gravity models, and applicable across different datasets and scales.

Conclusions:

• Machine-learned, closed-form models offer a powerful and interpretable approach to human mobility modeling.
• These models capture fundamental, universal features of human mobility, advancing urban planning and public health research.
• The findings suggest a new paradigm for modeling mobility flows that balances accuracy with interpretability.