CARE improves long-term forecasting with uncertainty signals
CARE: A Lightweight Plug-in Gated Correction and Uncertainty-aware Module for Long-term Time Series Forecasting
Machine Learning
Summary
Predicting things like electricity use or traffic flow far into the future is hard, and different parts of the forecast can be more or less reliable. The authors propose a small add-on called CARE that works alongside existing forecasting tools to correct their predictions and also estimate which predictions are more risky or uncertain. CARE does this without changing the original forecasting system’s structure and gives a clear signal about where the forecast might be wrong. This helps users know when to trust the forecast and when to be cautious.
What this means in practice
- •For energy demand planners: Improve long-term electricity load forecasts with an add-on that signals uncertain predictions to guide scheduling decisions.
- •For traffic management teams: Enhance traffic flow predictions by flagging times and locations with higher forecasting errors through CARE’s risk gating module.
Authors
Guo Cheng, Changlong Lv, Jingyi Hou
Abstract
Multivariate long-horizon forecasting is critical to electricity load scheduling and traffic flow management, and to financial risk control. Existing deterministic backbones output a single trajectory, masking heterogeneous prediction difficulty across horizons and channels and providing no localized reliability signal. We present CARE (Corrective branch with Aligned context and Relative-error Estimation), a lightweight plug-in that enhances any deterministic forecaster without architectural redesign. Operating in parallel with the base model, CARE resamples historical context to match the forecast horizon, learns residual correction patterns from this aligned history, and applies scale-aware bounded updates modulated by per-coordinate sigmoid risk gates. A multi-objective loss jointly optimizes forecast accuracy, residual tracking, risk alignment, and base-model anchoring. Across eight benchmarks with three representative backbones, CARE improves accuracy with marginal parameter and latency overhead. Its risk gates reliably identify high-error regions: on Weather, the highest-gate tertile exhibits nearly four times the error of the lowest-gate tertile, offering planners an interpretable per-step trust signal. Code is available at https://github.com/CG-BNYC/CARE.