This article introduces a selective forecasting framework that allows models to abstain from high-risk predictions by modeling the empirical percentile of forecasting errors through metalearning. By using scale-invariant statistics derived from recent lags, the method decouples rejection decisions from forecasts to enable transfer across heterogeneous time series.
- The framework models the empirical percentile of forecasting errors based on structural characteristics extracted from recent lags via metalearning.
- It utilizes a scale-invariant statistic that is domain-agnostic, allowing effective abstention transfer across different types of time series.
- Experiments in both in-domain and transfer learning settings demonstrate that rejecting challenging samples consistently improves forecasting accuracy across coverage levels.
This approach addresses the limitation of existing strategies tied to training domains, enabling more reliable performance on inherently difficult prediction instances.