Accurate carbon market regime detection is essential for risk management and policy evaluation, yet traditional econometric and machine-learning methods struggle with the policy-driven, non-stationary nature of carbon prices. This paper introduces the first systematic use of shapelet-based time-series classification for regime identification in the EU ETS, using 5,217 daily EUA futures prices (2005–2025). The method discovers 104 discriminative temporal patterns and achieves 99.68% test accuracy with an ARI of 0.850, outperforming standard baselines such as Gaussian Mixture Models (ARI 0.519), K-Means (0.435), and PELT change-point detection (0.676). A key advantage is interpretability: shapelets provide visual, economically meaningful temporal motifs that explain regime transitions. Statistical tests confirm significant improvements over all baselines ( \(p<0.001\) , Cohen’s \(d>2.0\) ). Overall, the results show that shapelet mining resolves the accuracy–interpretability tradeoff and offers a robust, transparent framework for regulatory time-series analysis.