Mixing individual and collective behaviors to predict out-of-routine mobility

Area of Science:

• Mobility studies
• Urban planning
• Computational social science

Background:

• Predicting human displacement is vital for urban design, traffic management, and epidemic control.
• Existing models often fail to accurately predict non-routine human movements.
• Individual mobility models struggle with emergent collective behaviors.

Purpose of the Study:

• To develop a novel approach for predicting human displacements by dynamically integrating individual and collective mobility patterns.
• To enhance the accuracy of mobility predictions, particularly for out-of-routine behaviors.
• To improve the robustness of predictive models during disruptive events.

Main Methods:

• Developed a model that dynamically integrates individual and collective mobility behaviors.
• Leveraged collective intelligence to improve prediction accuracy.
• Evaluated the model on millions of privacy-preserving trajectories across five US cities.

Main Results:

• The proposed model significantly outperforms advanced deep learning methods in predicting out-of-routine mobility.
• Model effectiveness is pronounced in urban areas with high point-of-interest density, where collective behaviors dominate.
• The model demonstrated sustained predictive capabilities during the COVID-19 pandemic, unlike individual-based models.

Conclusions:

• Integrating individual and collective mobility behaviors offers a more transparent and accurate prediction of human displacements.
• This approach enhances understanding and management of mobility challenges in urban environments.
• The model provides a robust framework for predicting mobility during both normal and disruptive conditions.