Delta-XAI: A Unified Framework for Explaining Prediction Changes in Online Time Series Monitoring

2025-11-28

AI Summary

Existing XAI methods for online time-series monitoring typically process individual time steps independently, neglecting temporal dependencies—leading to poor interpretability of dynamic predictions, limited online adaptability, and a lack of standardized evaluation frameworks. To address these limitations, we propose Delta-XAI, a unified framework for explainable online time-series monitoring. First, we introduce the first multi-dimensional XAI evaluation framework specifically designed for online time-series scenarios. Second, we propose SWING, a novel explanation method that integrates sliding windows with gradient path ensembling to explicitly model temporal dependencies, thereby enhancing explanation fidelity and temporal consistency. Third, we systematically adapt and rigorously evaluate classical gradient-based methods—including Integrated Gradients—demonstrating their superior performance over recent XAI approaches across multiple benchmark time-series datasets. All code is publicly available.

Technology Category

• Humans and AI: Explainable AI (XAI) for Human Understanding
• Machine Learning: Time-Series/Data Streams
• Computer Vision: Interpretability, Explainability, and Transparency

Application Category

• User Modeling, Personalization and Recommendation: Explainable and interpretable methods for personalization
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Explaining online time series monitoring models is crucial across sensitive domains such as healthcare and finance, where temporal and contextual prediction dynamics underpin critical decisions. While recent XAI methods have improved the explainability of time series models, they mostly analyze each time step independently, overlooking temporal dependencies. This results in further challenges: explaining prediction changes is non-trivial, methods fail to leverage online dynamics, and evaluation remains difficult. To address these challenges, we propose Delta-XAI, which adapts 14 existing XAI methods through a wrapper function and introduces a principled evaluation suite for the online setting, assessing diverse aspects, such as faithfulness, sufficiency, and coherence. Experiments reveal that classical gradient-based methods, such as Integrated Gradients (IG), can outperform recent approaches when adapted for temporal analysis. Building on this, we propose Shifted Window Integrated Gradients (SWING), which incorporates past observations in the integration path to systematically capture temporal dependencies and mitigate out-of-distribution effects. Extensive experiments consistently demonstrate the effectiveness of SWING across diverse settings with respect to diverse metrics. Our code is publicly available at https://anonymous.4open.science/r/Delta-XAI.

Problem

• Explaining prediction changes in online time series monitoring
• Addressing overlooked temporal dependencies in XAI methods
• Providing a unified framework and evaluation for online settings

Innovation

• Adapts existing XAI methods via wrapper function
• Introduces principled evaluation suite for online setting
• Proposes SWING method capturing temporal dependencies systematically