A Systematic Review on Explainable AI for Time Series Classification

Explainable AI (XAI) for time series classification has gained attention as AI enters safety-critical domains like healthcare, industry, and finance, which routinely generate time series data whose temporal and multivariate dependencies pose distinct challenges for explanations. This systematic review analyzes 219 publications from 2014 to 2025 across ten categories spanning data, models, explanations, evaluation, and computational cost. Results reveal a dominance of local post-hoc subsequence-based explanations, while ante-hoc architectures, global explanations, and channel-aware attribution remain under-explored. A cross-category analysis identifies challenges and directions for future research. Frequency-aware and counterfactual approaches emerge as notable directions, while formal evaluation remains largely absent.