ARTFEED — Contemporary Art Intelligence

C-SHAP: Concept-Based XAI for Time Series

other · 2026-04-25

A new explainable AI method called C-SHAP addresses the lack of high-level temporal explanations in time series analysis. Unlike existing techniques that focus on points or subsequences, C-SHAP uses concepts defined as high-level patterns extracted from the data. It leverages the SHAP method to measure each concept's influence on predictions. The approach targets high-stakes domains like healthcare and industry where model reliability is critical. The paper is available on arXiv under ID 2504.11159.

Key facts

  • C-SHAP is a concept-based XAI approach for time series
  • Concepts are high-level patterns extracted from time series data
  • C-SHAP uses SHAP to determine concept influence on predictions
  • Targets high-stakes domains like healthcare and industry
  • Existing XAI techniques for time series are point- or subsequence-based
  • Paper published on arXiv with ID 2504.11159

Entities

Institutions

  • arXiv

Sources