Urban Anomalies: A Simulated Human Mobility Dataset with Injected Anomalies

Sumamry:

"Urban Anomalies" introduces a large-scale simulated dataset designed to aid in detecting anomalies in human mobility patterns. Built on the Patterns of Life Simulation, the dataset captures deviations from typical human behaviors to simulate real-world anomalies, addressing data scarcity and privacy issues common in human mobility research.

Methodology:

The study defines four anomaly types (hunger, social, work, and interest) with varying intensity levels (yellow, orange, red). Anomalous behaviors are simulated through three methods: random central injection, infectious disease spread, and location-based spread. The dataset comprises labeled trajectories and social connections for agents in simulated urban environments (e.g., Atlanta and Berlin), with distinct training and testing phases to evaluate anomaly detection models.

Results:

The dataset features extensive anomaly scenarios, including subtle and pronounced deviations from normal patterns. Agents display varied visitation, social interaction, and work patterns depending on anomaly type and intensity. The infection-based models create epidemiological-like spread patterns, showing scalability and realistic anomaly progression.

Conclusion:

This dataset provides a valuable tool for developing and testing algorithms for anomaly detection in human mobility data. The open access to the dataset allows researchers to enhance models with complex, realistic anomaly scenarios, enabling breakthroughs in public health, urban safety, and behavioral analysis applications.

Full Paper: