![[2509.20936] GenFacts-Generative Counterfactual Explanations for Multi-Variate Time Series](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2509.20936] GenFacts-Generative Counterfactual Explanations for Multi-Variate Time Series
Summary
The paper introduces GenFacts, a generative framework for creating counterfactual explanations in multivariate time series, improving model transparency and interpretability.
Why It Matters
Understanding how models make predictions is crucial in AI, especially in sensitive applications. GenFacts enhances the quality of counterfactual explanations, making them more plausible and user-friendly, which can lead to better decision-making in various fields, including healthcare and finance.
Key Takeaways
- GenFacts uses a class-discriminative variational autoencoder for generating counterfactuals.
- It outperforms existing methods by 18.7% in plausibility and achieves high interpretability scores.
- The framework emphasizes user-centered interpretability over mere sparsity in counterfactuals.
- Evaluated on radar gesture data and handwritten letter trajectories, showing versatility.
- Highlights the importance of realistic counterfactuals in actionable insights for time series data.
Computer Science > Machine Learning arXiv:2509.20936 (cs) [Submitted on 25 Sep 2025 (v1), last revised 17 Feb 2026 (this version, v3)] Title:GenFacts-Generative Counterfactual Explanations for Multi-Variate Time Series Authors:Sarah Seifi, Anass Ibrahimi, Tobias Sukianto, Cecilia Carbonelli, Lorenzo Servadei, Robert Wille View a PDF of the paper titled GenFacts-Generative Counterfactual Explanations for Multi-Variate Time Series, by Sarah Seifi and 5 other authors View PDF HTML (experimental) Abstract:Counterfactual explanations aim to enhance model transparency by showing how inputs can be minimally altered to change predictions. For multivariate time series, existing methods often generate counterfactuals that are invalid, implausible, or unintuitive. We introduce GenFacts, a generative framework based on a class-discriminative variational autoencoder. It integrates contrastive and classification-consistency objectives, prototype-based initialization, and realism-constrained optimization. We evaluate GenFacts on radar gesture data as an industrial use case and handwritten letter trajectories as an intuitive benchmark. Across both datasets, GenFacts outperforms state-of-the-art baselines in plausibility (+18.7%) and achieves the highest interpretability scores in a human study. These results highlight that plausibility and user-centered interpretability, rather than sparsity alone, are key to actionable counterfactuals in time series data. Comments: Subjects: Machine Lear...
