SpatialEpiBench: Benchmarking Spatial Information and Epidemic Priors in Forecasting
SpatialEpiBench: Benchmarking Spatial Information and Epidemic Priors in Forecasting
Authors: Ruiqi Lyu, Alistair Turcan, Bryan Wilder Date: 2026-05-07 Paper ID: openalex:2605.06530
Summary
SpatialEpiBench addresses the lack of standardized evaluation frameworks in epidemic forecasting by providing a benchmark with 11 datasets and rolling evaluation protocols that reflect real-time public health practice. The study evaluates diverse models incorporating geographic adjacency and epidemiological priors, revealing a significant performance gap compared to simple persistence baselines. The authors categorize model failures into poor outbreak anticipation, sensitivity to data noise, and ineffective use of spatial graph structures.
Key Contributions
- Introduces SpatialEpiBench, a benchmark for spatiotemporal epidemic forecasting across 11 datasets using realistic rolling evaluation protocols.
- Demonstrates that existing adjacency-informed models and epidemic priors frequently underperform simple last-value baselines across various forecasting horizons.
- Identifies three critical failure modes in current models: poor outbreak anticipation, limited robustness to data sparsity/noise, and weak utility of simple geographic adjacency for epidemiological modeling.
Open Questions & Future Work
Key Concepts
- spatialepibench: A benchmark for spatiotemporal epidemic forecasting incorporating 11 datasets and standardized rolling evaluations.
Archivist Review
I approved the benchmark as a foundational contribution to standardized evaluation in the domain and the open question regarding the validity of geographic priors in epidemiology, as these address significant, recurring issues in spatiotemporal forecasting. I rejected the request to label the 11 datasets individually as they are aggregate components of the benchmark, adhering to the scarcity policy for datasets.
Approved Concepts
- SpatialEpiBench: It is a primary, standardized benchmark for spatiotemporal epidemic forecasting designed to reflect realistic public-health practices.
Approved Open Questions
- Utility of Geographic Adjacency: This is a fundamental failure mode in the field that limits the effectiveness of current spatiotemporal modeling approaches in public health.
Links
Metadata & Links
- url
- https://arxiv.org/abs/2605.06530
- paper_id
- 2605.06530
- paper_source
- openalex
- domain
- time-series
- tags
- time-series-forecastingepidemic-forecastingspatiotemporal-modelingbenchmarking
- architectures
-
- datasets
-
- concept_slugs
- spatialepibench
- dataset_slugs
-
- skill
- TimeSeriesSkill
- created_at
- 2026-05-10T06:04:03Z