Research map / Multi-agent systems and games / Human-AI and goal-conditioned
A Generalizable Theory-Driven Agent-Based Framework to Study Conflict-Induced Forced Migration
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes a theory-driven agent-based model of conflict-induced forced migration (Theory of Planned Behavior, spatiotemporal gravity risk model, bi-threshold synchronous dynamical system over household network), calibrated on Ukraine 2022 and transferred to Burundi 2015.
Problem
Existing forced-migration models focus on destination countries, are country-specific, or ignore social theory and agent-agent interaction, so they cannot explain which people in a conflict-affected origin region flee and when.
Approach
Combines the Theory of Planned Behavior with a spatiotemporal gravity model of perceived risk from conflict events and a bi-threshold rule over a household graph dynamical system, so migration decisions depend on both agent-event and agent-agent interaction at household level.
Key results
- On Ukraine refugee estimation, the ABM achieves MAPE 20.72% vs 74.95% for Pandey et al., with PCC 0.91 vs 0.87.
- Top-5 IDP origin oblast match with IOM is 80%, 80%, and 100% across three survey rounds for the ABM, versus 60%, 80%, and 60% for Pandey et al.
- The model is transferred without re-calibration to the 2015 Burundi conflict and produces estimates close to observed cumulative refugee data.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use the ABM framework instead of Pandey et al. | Ukraine refugee estimation · Ukraine | MAPE: 74.95%±24.27% -> 20.72% ± 3.22% | Helped |
| Use the ABM framework instead of Pandey et al. | Ukraine refugee estimation · Ukraine | RMSE: 1.06e5 ± 5.17e4 -> 2.6e4 ± 1.3e3 | Helped |
| Use the ABM framework instead of Pandey et al. | Ukraine refugee estimation · Ukraine | PCC: 0.87 ± 0.01 -> 0.91 ± 0.01 | Helped |
| Use the ABM framework instead of Pandey et al. | IDP-origin estimation · Ukraine, Round 1 | Top-5 match: 60% -> 80% | Helped |
| Use the ABM framework instead of Pandey et al. | IDP-origin estimation · Ukraine, Round 2 | Top-5 match: 80% -> 80% | No clear effect |
| Use the ABM framework instead of Pandey et al. | IDP-origin estimation · Ukraine, Round 3 | Top-5 match: 60% -> 100% | Helped |
| Enforce entrapment conditions in selected raions, then relax them | ABM · Ukraine, Polohivskyi raion | refugee outflow | Mixed |
Surprising findings
- Entrapment scenario shows that after restrictions are lifted, the refugee surge is almost twice the surge seen in the status quo scenario.
Limitations
- Agent locations are assumed fixed to household locations across all timesteps, since real-time individual location data is unavailable.
- Only age and gender demographics are considered; no income or other features.
- Return migration is not modeled, limiting longer time horizons.
Experimental setup
Datasets: Synthetic population (Biocomplexity Institute) ACLED conflict event data HUMDATA border crossing data IOM IDP survey data Burundi 2015 observed data from Suleimenova et al. 2017
Benchmarks: HUMDATA border crossings (Ukraine 2022) IOM IDP survey rounds Burundi 2015 cumulative refugee data
Metrics: MAPE RMSE PCC top-5 oblast match percentage
Baselines: Pandey et al. 2023
Builds on: Theory of Planned Behavior Weidmann spatiotemporal gravity model Bi-threshold model Graph dynamical system Pandey et al. 2023 exponential decay model
Closest papers by meaning
- Generating Origin-Destination Matrices in Neural Spatial Interaction Models (NeurIPS 2024 · similarity 0.80)
- MF-LLM: Simulating Population Decision Dynamics via a Mean-Field Large Language Model Framework (NeurIPS 2025 · similarity 0.80)
- World Models Should Prioritize the Unification of Physical and Social Dynamics (NeurIPS 2025 · similarity 0.78)
- Can LLM Agents Respond to Disasters? Benchmarking Heterogeneous Geospatial Reasoning in Emergency Operations (NeurIPS 2026 · similarity 0.78)
- Beyond Prediction: Steering VLM Agents with Retrospective World Modeling (NeurIPS 2026 · similarity 0.78)
- ROK-FORTRESS: Measuring the Effect of Geopolitical Transcreation for National Security and Public Safety (NeurIPS 2026 · similarity 0.77)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.