Time series forecast steps explained with step-specific past value maps
Explaining Time Series Forecasting with Horizon-Resolved Attribution
Machine LearningArtificial Intelligence
Summary
Time series models predict future values based on past data, but current explanation methods assume all future steps depend on the same past points. The authors found this is not true: different future moments rely on different past data. They introduced a method called Horizon-Resolved eXplanation (HRX) that shows which past inputs influence each forecast step separately. Their approach works with existing models and improves understanding of forecasts by highlighting step-wise dependencies. They also show these dependencies can be summarized with a small number of shared explanation maps.
What this means in practice
- •For financial analysts: Identify which past financial metrics influence different future forecast steps to improve interpretation of market predictions.
- •For energy demand planners: Clarify step-specific input influences on energy use forecasts to support better operational decision-making.
Authors
Seunghan Lee, Jun Seo, Jaehoon Lee, Junhyeok Kang, Sangjun Han, Sungdong Yoo, Minjae Kim, Tae Yoon Lim, Dongwan Kang, Hwanil Choi, Soonyoung Lee, Wonbin Ahn
Abstract
Recent advances in explaining time series (TS) models have produced methods that identify which past values a prediction depends on. However, most existing methods return a single importance vector, assuming that every predicted step depends on the same past values. In this paper, we show that this assumption does not hold, as different forecast steps depend on different past values. Motivated by this observation, we propose Horizon-Resolved eXplanation (HRX), which adds a horizon axis to the explanation, so that every forecast step receives its own importance map. HRX is a simple yet effective plug-in framework with three components: 1) an estimator that reads these maps out of any differentiable forecaster without modifying the TS backbone, 2) an evaluation protocol that validates the horizon axis by measuring how much a single forecast step changes when the inputs an importance map ranks highest are removed, and 3) a rank criterion that predicts in advance whether the axis is worth resolving on a given TS. We further show that this step-wise dependence is low-dimensional, as the explanations of all steps are built from a few shared maps whose number does not grow with the forecast length. Extensive experiments across various backbones and datasets show that the improvement comes from the horizon axis and holds for estimators of previous explanation methods. Code is available at https://github.com/seunghan96/HRX.