Delta-XAI framework enhances time series model explainability with temporal focus

Researchers have introduced Delta-XAI, a new framework designed to explain changes in predictions made by online time series monitoring models. This framework addresses limitations in existing methods that often analyze time steps independently, failing to capture crucial temporal dependencies. Delta-XAI adapts 14 existing explainability techniques and includes a new evaluation suite for online settings, demonstrating that adapted gradient-based methods like Integrated Gradients can be effective. The paper also proposes Shifted Window Integrated Gradients (SWING) to systematically incorporate past observations for improved temporal analysis. AI

Summary written by gemini-2.5-flash-lite from 1 source. How we write summaries →

IMPACT Introduces a novel framework for improving the explainability of time series models, potentially enhancing trust and adoption in sensitive domains.

RANK_REASON The cluster contains an academic paper detailing a new framework for explaining AI model predictions.

Read on arXiv cs.LG →

• Delta-XAI
• Integrated Gradients
• SWING
• arXiv
• Changhun Kim

• paper
• other