Mobility Anomaly Generation

Mobility Anomaly Generation produces realistic, controllable deviations from normal movement while preserving causal explanations and physical feasibility.

Overview

Anomalies in real mobility data are rare, heterogeneous, and difficult to label. This project generates anomalies by changing the behavioral cause of movement, then enforcing spatial and kinematic constraints on the resulting trajectory. The approach supports experiments in which anomaly type, intensity, duration, and context are known, enabling more rigorous evaluation than arbitrary coordinate perturbations.

Research Questions

• What behavioral causes produce meaningful mobility anomalies?
• How can anomalous traces remain physically and geographically feasible?
• How should detectors be evaluated when anomalies vary in cause, duration, and severity?

Methods

• LLM-driven behavioral scenario generation
• Kinematic and network constraints
• Causally grounded anomaly taxonomies
• Controlled benchmarking of anomaly detectors

Applications

• Public safety and infrastructure monitoring
• Transportation-system diagnostics
• Evaluation of GeoAI anomaly detectors
• Synthetic rare-event data generation

Selected Outputs

• Grounded Anomalies framework
• LLM-driven behavior with kinematic constraints
• Benchmark-ready anomalous mobility traces