Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Physics-Guided Counterfactual Explanations for Large-Scale Multivariate Time Series: Application in Scalable and Interpretable SEP Event Prediction

Pranjal Patil, Anli Ji, Berkay Aydin

TL;DR

This work addresses the interpretability gap in large-scale MVTS SEP forecasting by introducing Physics-Guided Counterfactual Explanations (PGCE) that enforce physical plausibility through flux-ordering, channel-wise bounds, and temporal smoothness. The method extends DiCE with a constrained genetic explainer and a local-global reconstruction pipeline to map perturbations into physically realizable time series, enabling faithful and actionable explanations. Empirical results on GOES-based SEP data show substantial reductions in $DTW$ distance, improved sparsity, and reduced runtime compared with standard DiCE, while maintaining perfect fidelity to the underlying classifier. The framework thus provides scalable, physics-consistent counterfactual explanations that enhance trust, visualization, and scientific insight for space weather forecasting and potentially other sequential, domain-constrained big-data applications.

Abstract

Accurate prediction of solar energetic particle events is vital for safeguarding satellites, astronauts, and space-based infrastructure. Modern space weather monitoring generates massive volumes of high-frequency, multivariate time series (MVTS) data from sources such as the Geostationary perational Environmental Satellites (GOES). Machine learning (ML) models trained on this data show strong predictive power, but most existing methods overlook domain-specific feasibility constraints. Counterfactual explanations have emerged as a key tool for improving model interpretability, yet existing approaches rarely enforce physical plausibility. This work introduces a Physics-Guided Counterfactual Explanation framework, a novel method for generating counterfactual explanations in time series classification tasks that remain consistent with underlying physical principles. Applied to solar energetic particles (SEP) forecasting, this framework achieves over 80% reduction in Dynamic Time Warping (DTW) distance increasing the proximity, produces counterfactual explanations with higher sparsity, and reduces runtime by nearly 50% compared to state-of-the-art baselines such as DiCE. Beyond numerical improvements, this framework ensures that generated counterfactual explanations are physically plausible and actionable in scientific domains. In summary, the framework generates counterfactual explanations that are both valid and physically consistent, while laying the foundation for scalable counterfactual generation in big data environments.

Physics-Guided Counterfactual Explanations for Large-Scale Multivariate Time Series: Application in Scalable and Interpretable SEP Event Prediction

TL;DR

This work addresses the interpretability gap in large-scale MVTS SEP forecasting by introducing Physics-Guided Counterfactual Explanations (PGCE) that enforce physical plausibility through flux-ordering, channel-wise bounds, and temporal smoothness. The method extends DiCE with a constrained genetic explainer and a local-global reconstruction pipeline to map perturbations into physically realizable time series, enabling faithful and actionable explanations. Empirical results on GOES-based SEP data show substantial reductions in distance, improved sparsity, and reduced runtime compared with standard DiCE, while maintaining perfect fidelity to the underlying classifier. The framework thus provides scalable, physics-consistent counterfactual explanations that enhance trust, visualization, and scientific insight for space weather forecasting and potentially other sequential, domain-constrained big-data applications.

Abstract

Accurate prediction of solar energetic particle events is vital for safeguarding satellites, astronauts, and space-based infrastructure. Modern space weather monitoring generates massive volumes of high-frequency, multivariate time series (MVTS) data from sources such as the Geostationary perational Environmental Satellites (GOES). Machine learning (ML) models trained on this data show strong predictive power, but most existing methods overlook domain-specific feasibility constraints. Counterfactual explanations have emerged as a key tool for improving model interpretability, yet existing approaches rarely enforce physical plausibility. This work introduces a Physics-Guided Counterfactual Explanation framework, a novel method for generating counterfactual explanations in time series classification tasks that remain consistent with underlying physical principles. Applied to solar energetic particles (SEP) forecasting, this framework achieves over 80% reduction in Dynamic Time Warping (DTW) distance increasing the proximity, produces counterfactual explanations with higher sparsity, and reduces runtime by nearly 50% compared to state-of-the-art baselines such as DiCE. Beyond numerical improvements, this framework ensures that generated counterfactual explanations are physically plausible and actionable in scientific domains. In summary, the framework generates counterfactual explanations that are both valid and physically consistent, while laying the foundation for scalable counterfactual generation in big data environments.
Paper Structure (21 sections, 8 equations, 4 figures, 1 table, 2 algorithms)

This paper contains 21 sections, 8 equations, 4 figures, 1 table, 2 algorithms.

Table of Contents

  1. Introduction
  2. Related Work
  3. Methodology
  4. Problem Formulation
  5. Dataset Details
  6. Evaluation of Candidate Models and Baseline Selection
  7. Explainer Model to Generate DiCE Based Counterfactual Explanations with Physical Constraints
  8. Reconstruction of Timeseries Using Local-Global Offsets for the Counterfactual Explanation of a SEP timeseries
  9. Experiments
  10. Experiment Design
  11. Evaluation Metrics
  12. Experiment 1: Physics-guided constraint optimization for explainer model
  13. Experiment 2: Comparative Study of Physics-Guided Counterfactual Explanations Against Standard DiCE
  14. Experiment 3: Fidelity Evaluation of Counterfactual
  15. Results and Discussions
  16. ...and 6 more sections

Figures (4)

  • Figure 1: GOES-SEM Solar Proton Flux and their Counterfactual Explanation (CFE) with Blended Overlapping Windows
  • Figure 2: Illustration of diverse counterfactual reconstructions for a single event. The original input corresponds to a non-SEP event, while the generated counterfactual explanations reflect alternative scenarios classified as SEP events.
  • Figure 3: Effect of Hyperparameter Settings on Generation of Counterfactual Explanations' Quality Metrics
  • Figure 4: Comparison of Physics-Guided Counterfactual Explanations (PGCE) Framework and DiCE on Proximity, Sparsity, and Diversity (normalized to [0,1]).