Automated Exogenous Group Discovery
Background: The selection and integration of exogenous variables in forecasting models is often limited by reliance on expert-defined domain knowledge for grouping.
Question / Future Work: Research is needed into automated, data-driven group discovery mechanisms—such as attention-based clustering or learned sparsity patterns—to eliminate the need for manual site-specific configuration and improve model adaptability.
Why It Matters: Automated grouping is crucial for reducing engineering effort and improving the generalizability of forecasting models.
Evidence: the physical grouping prior, while effective, currently relies on domain knowledge to define the variable groups; automating this grouping through data-driven group discovery could improve applicability to domains where physical grouping is less obvious.
Metadata & Links
- created_at
- 2026-05-15T06:15:03Z