Explainable Artificial Intelligence (XAI) on TimeSeries Data: A Survey

Time series AI systems lack transparency compared to other AI domains: The survey highlights that explainability methods for time series data have received significantly less attention than those developed for computer vision or natural language processing, leaving a critical gap in interpretability for this important class of models.

Deep learning models dominate time series applications but are often "black boxes": Most state-of-the-art methods applied to time series use complex deep learning approaches that are too difficult to interpret, creating a major drawback for real-world deployment in critical applications.

Critical applications demand explainability: The paper emphasizes that high-stakes domains like medical diagnosis and autonomous driving require interpretable AI systems, making the lack of XAI methods for time series a significant concern for responsible deployment in these fields.

Explanations can build trust and confidence in AI systems: The survey reflects on how explanation methods can help establish confidence and trust in AI systems, suggesting that making time series models more interpretable is not just a technical requirement but essential for responsible AI adoption.

Multiple XAI approaches exist for time series: The paper presents an overview of existing explainable AI methods applied to time series and illustrates the different types of explanations they can produce, indicating that solutions are available but underutilized compared to other domains.