Information Bottleneck Learning for Faithful Time Series Forecasting Explanations

2026/07/30 by Xu Zheng, Wei Cheng, Zhuomin Chen +3
Computer Science · #cs.LG #cs.AI

paper · pdf

17 pages, 6 figures, 8 tables

arxiv created 2026/07/30 · arxiv updated 2026/07/31

Abstract

As forecasts increasingly drive decisions in fields such as energy, transportation, and healthcare, understanding the historical data behind these predictions has become as crucial as the predictions themselves. Although existing interpretable-by-design forecasters reveal their internal structures, they offer no guarantee that these structures faithfully reflect the underlying evidence driving the predictions. In contrast, while faithfulness-oriented methods explicitly verify model behavior, they are almost exclusively designed for post-hoc classification tasks. To bridge this gap, we propose IB-Forecast, an inherently interpretable multivariate time-series forecasting framework. It decomposes forecasting into a learned periodic component and a residual component computed with explainable masks over input tokens. With a budget-constrained information bottleneck, end-to-end optimization enables users to directly control explanation sparsity. With a rigorous faithfulness evaluation protocol, extensive experiments demonstrate that IB-Forecast matches the forecasting error of leading black-box models while providing faithful explanations at no additional inference cost. Furthermore, under a matched sparsity budget, these native explanations consistently surpass gradient-based, occlusion-based, and optimization-based baselines across all evaluated datasets. Ultimately, whereas the native explanations of existing interpretable forecasters exhibit poor faithfulness, IB-Forecast guarantees high explanation fidelity, requiring only 14-20% of the observations to deliver low-error predictions.

Citations

• TIMING: Temporality-Aware Integrated Gradients for Time Series Explanation
• Interpretability for Time Series Transformers using A Concept Bottleneck Framework
• F-Fidelity: A Robust Framework for Faithfulness Evaluation of Explainable AI
• CycleNet: Enhancing Time Series Forecasting through Modeling Periodic Patterns
• TimeX++: Learning Time-Series Explanations with Information Bottleneck
• Explaining Time Series via Contrastive and Locally Sparse Perturbations
• Encoding Time-Series Explanations through Self-Supervised Model Behavior Consistency
• Learning Perturbations to Explain Time Series Predictions
• A Time Series is Worth 64 Words: Long-term Forecasting with Transformers
• Interpretable and Generalizable Graph Learning via Stochastic Attention Mechanism
• Explaining Time Series Predictions with Dynamic Masks
• Informer: Beyond Efficient Transformer for Long Sequence Time-Series Forecasting
• Informer: Beyond Efficient Transformer for Long Sequence Time-Series Forecasting
• What Makes for Good Views for Contrastive Learning?
• Time-series forecasting with deep learning: a survey
• WeatherBench: A Benchmark Data Set for Data‐Driven Weather Forecasting
• Temporal Fusion Transformers for Interpretable Multi-horizon Time Series\n Forecasting
• Temporal Fusion Transformers for interpretable multi-horizon time series forecasting
• ERASER: A Benchmark to Evaluate Rationalized NLP Models
• N-BEATS: Neural basis expansion analysis for interpretable time series forecasting
• Attention is not Explanation
• Learning Sparse Neural Networks through L₀ Regularization
• Axiomatic Attribution for Deep Networks
• The Concrete Distribution: A Continuous Relaxation of Discrete Random Variables
• Deep Inside Convolutional Networks: Visualising Image Classification Models and Saliency Maps
• The information bottleneck method

Related

• Time Series Forecasting as Reasoning: A Slow-Thinking Approach with Reinforced LLMs
• Faithful and Interpretable Explanations for Complex Ensemble Time Series Forecasts using Surrogate Models and Forecastability Analysis
• TsSHAP: Robust model agnostic feature-based explainability for time series forecasting
• Abstain Mask Retain Core: Time Series Prediction by Adaptive Masking Loss with Representation Consistency
• Counterfactual Explanations for Time Series Forecasting