Selective Time Series Forecasting via Metalearning

2026-06-22

AI Summary

This work addresses the challenge in time series forecasting where certain samples are inherently difficult to predict accurately, and existing rejection mechanisms rely on domain-specific information, limiting their generalization. The paper proposes a domain-agnostic selective prediction framework that, for the first time, integrates meta-learning with the lagged structural characteristics of time series. By modeling empirical percentiles of prediction errors, the approach enables decoupled rejection decisions without requiring task-specific tuning. It supports transfer across heterogeneous time series and effectively abstains from predicting high-difficulty samples both within-domain and under distribution shifts, significantly improving predictive accuracy across varying coverage levels.

Abstract

Deep learning methods have achieved state-of-the-art in time series forecasting, yet their accuracy varies considerably across samples, as some instances remain inherently difficult to predict. Reject option mechanisms, which allow models to abstain from high-risk predictions, are well established in classification and regression but underexplored in forecasting. Existing abstention strategies typically rely on proxies, such as the width of the prediction interval or learned confidence scores derived from forecasts. However, these approaches are inherently tied to the training domain, limiting their ability to generalize. We propose a selective forecasting framework that addresses this limitation by modeling the empirical percentile of forecasting errors, that is, a scale-invariant statistic, based on structural characteristics extracted from recent lags via metalearning. By decoupling the rejection decision from the forecast itself and grounding it in domain-agnostic features, the framework enables effective abstention transfer across heterogeneous time series. Experiments in both in-domain and transfer learning settings show that rejecting samples predicted as challenging consistently improves forecasting accuracy across coverage levels.

Problem

• selective forecasting
• reject option
• time series forecasting
• abstention
• generalization

Innovation

• selective forecasting
• metalearning
• abstention mechanism
• domain-agnostic features
• forecasting error percentile