Software Frameworks for Explainable AI in Time Series Classification: A Systematic Review

Most XAI tools for time series aren't purpose-built: The review identifies only six software frameworks that explicitly support time series classification, suggesting that practitioners often must adapt general-purpose AI explanation tools rather than use specialized solutions designed for their specific needs.

Critical evaluation gaps exist: Only two evaluation metrics have been developed specifically for assessing explanations in time series contexts, meaning current tools may not adequately measure whether explanations are actually meaningful or reliable for this type of data.

Frequency-domain explanations are largely missing: Despite their relevance to time series analysis, only one framework supports frequency-domain explanations, indicating a significant gap between what would be scientifically useful and what tools currently provide to users.

The same explanation method produces inconsistent results across frameworks: Identical XAI methods yield different outputs depending on which software framework is used, raising concerns about reproducibility and the reliability of explanations in real-world decision-making applications.

Reproducibility and transparency remain underdeveloped: The fragmented state of XAI research for time series means practitioners lack standardized benchmarking support and clear documentation, making it difficult to verify that AI systems are truly trustworthy or to compare different explanation approaches fairly.