Explaining Time Series Classifiers Through Post-Hoc XAI Methods Capturing Temporal Dependencies

Time series classification is essential in domains such as healthcare and finance, where accurate predictions can have significant real-world consequences. However, in many high-stakes applications, understanding why a model makes a certain decision is just as important as the prediction itself. While deep learning models excel at capturing complex temporal patterns, their black-box nature limits transparency, making it difficult to trust and interpret their decisions. Although eXplainable AI (XAI) methods have advanced considerably for image and tabular data, applying them to time series remains challenging due to the intricate temporal dependencies and high dimensionality of the data. Post-hoc model-agnostic XAI techniques offer a promising solution by providing explanations without altering the underlying model. This research focuses on developing novel post-hoc model-agnostic XAI methods specifically for time series classifiers. By elucidating prediction processes while preserving temporal structures, these methods seek to enhance interpretability and trust, thereby facilitating informed decision-making in high-stakes applications.