New TempShap framework offers detailed feature attribution for time-series forecasting

Researchers have introduced TempShap, a novel framework for explaining time-series forecasting models. Unlike existing methods that treat features as single entities, TempShap analyzes individual input cells across the time window and forecast horizon. This approach provides detailed profiles of when a feature is influential and how its impact changes over time. The framework is validated through planted temporal structures, exact cell games, and a closed-form attribution oracle, demonstrating its effectiveness in practical applications and across various forecasting benchmarks and model families. AI

IMPACT Enhances interpretability of time-series models, potentially improving debugging and trust in AI-driven forecasts.

RANK_REASON The cluster describes a new research paper introducing a novel method for explaining AI models. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• arXiv
• Hugging Face
• Shapley Values
• TempShap
• Time-Series Forecasting

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →