AI Science Tracker
EN

Petrozavodsk State University

1 discovery from 2026

Discoveries

Learned forecasts distinguish directional search and COVID-data relationships

Source check: Source checked. Not independent scientific confirmation.

Finding: Matched LightGBM forecasts of eight Google-search and COVID-related series found that opposite directions could differ in both prediction error and error-distribution behavior. For US loss-of-smell searches and reported cases, adding search history gave a 9.9% equivalent mean RMSE reduction, while the reverse direction changed error-distribution descriptors without improving mean accuracy. These are model-dependent observational patterns, not causal effects or a validated real-time warning system.

Technology: LightGBM gradient-boosted regression trees

Publication: Preprint · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.

AI contribution
Learned tree ensembles generate matched conditional forecasts whose held-out errors reveal the source- and direction-specific contrast; FEDE itself is a statistical descriptor, not a separate learned model.
Test and result
Each 400-day window used 14-day histories, one- to fourteen-day horizons and chronological 223-origin training/150-origin evaluation. Target-only and source-augmented branches shared settings. Results Tables 6 and 9 separate mean log-RMSE and forecast-error distribution entropy (FEDE).
Limitation / still unknown
Adjacent windows share 399 days and are not independent replications. Window/history settings were chosen sequentially for distributional stability. Relative Google Trends sampling, query wording, common media/seasonal/reporting effects and the fixed model class limit interpretation. No temporally constrained surrogate-source null test was included. Similar broad lead-lag findings were known; the represented refinement is the matched metric-dependent directional contrast, not the discovery that searches can precede cases.
Evidence sources and annotation context

Evidence context (descriptions follow the cited sources; test type is a catalog classification):

Evidence sources: https://arxiv.org/html/2610.28563v1 · https://arxiv.org/abs/2610.28563v1

Source checked · editorial details

The source page was checked against this entry. Independent confirmation is not assessed by this label.

Source type: paper. Paper: linked separately.

Review date: 21 Nov 2026. Review scope: Original v1 front, computational Methods, application Results and limitations; no independent reproduction.

No follow-up recorded; this is not a claim that no follow-up exists. Follow-up review scope and date are not recorded. No source-supported relationships recorded.

Editorial assessment: Impact 1/5 · Validation 2/5 · Novelty 2/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.

Open the interactive profile for the timeline, or browse all organizations and model families.