Post-hoc Attribution Rules (PHAR) is an integrated framework for enhancing the explainability of time series classification models. It transforms numerical feature importance derived from existing post-hoc, instance-wise explanation techniques (e.g., LIME, SHAP) into structured rules that are easily understood by humans. These rules enhance model transparency by defining interpretable intervals indicating where and when critical decision boundaries occur. PHAR performs similarly to existing rule-based methods, such as Anchor, while scaling more efficiently to long time series sequences and achieving broader instance coverage. A dedicated rule fusion step, which integrates rule sets using strategies such as weight selection and Lasso-based refinement, balances key quality metrics such as coverage, confidence, and simplicity. This fusion provides concise, unambiguous rules for each instance, improving both explanation fidelity and consistency. It also introduces a visualization technique that illustrates the specificity-generalization tradeoff in the derived rules. PHAR resolves the conflicting and overlapping explanations common to the Rashomon phenomenon with consistent, domain-specific insights. Comprehensive experiments on the UCR/UEA time series classification archive demonstrate that PHAR improves interpretability, decision transparency, and practical applicability for time series classification tasks.
