Towards a Comprehensive Solution for Generating Synthetic Human Mobility Data from Limited Input Data

Ronald H Davies, Rajiv Ramnath

Keywords: synthetic data, human mobility, generative, agents, privacy

TL;DR: This paper presents a novel ensemble approach to generating realistic synthetic mobility data without input data for nearly any well-mapped region.

Abstract: Human mobility data is essential for transportation planning, urban analytics, and public health, yet access to real-world traces is increasingly restricted due to privacy concerns and institutional controls. Synthetic mobility data offers a viable alternative, but many existing approaches require extensive data inputs or model retraining, raising barriers to reproducibility and privacy assurance. We present a data-minimal framework for generating synthetic human mobility trajectories without additional model training and using only two external sources. The method comprises three conceptual components: (1) profile generation to define agent attributes, (2) agent-based decision-making to plan full-day activity sequences, and (3) route and waypoint visualization to produce structured trajectories and geographic renderings. This design enforces semantic coherence and temporal feasibility while preserving privacy by avoiding real user data. The result is a reproducible, extensible approach for creating synthetic mobility datasets suitable for research and policy analysis under stringent data constraints

Primary Area: other topics in machine learning (i.e., none of the above)

Submission Number: 23910