<p>Accurate and interpretable arrhythmia classification from ECG signals is crucial for timely diagnosis and treatment of cardiac conditions. However, existing methods often struggle with capturing complex ECG patterns while lacking transparency in decision-making. To address these challenges, we propose Arrhythmia Classification with Interpretative Temporal Convolutional Transformer (AC-ITCT). A key innovation of our approach is the incorporation of a Temporal Convolutional Transformer (TCT) with a time attention mechanism, which effectively captures critical temporal dependencies in ECG signals, essential for accurate classification. Another significant advancement is the adaptive temporal resolution adjustment module, which dynamically adjusts the analysis of granularity across time, allowing the model to selectively focus on relevant ECG segments at varying time scales. The methodology also integrates an active learning strategy, enabling iterative model improvement by identifying the most informative data points for manual annotation, thus reducing the costs associated with data labeling. Additionally, an aggregation module combines relevant features across time steps, ensuring a comprehensive representation of ECG signals. To provide interpretability, an attention visualization mechanism is incorporated to highlight critical ECG segments that influence classification decisions. The AC-ITCT is evaluated on the MIT-BIH Arrhythmia Database, achieving 99.96% accuracy, 99.7% precision, 99.8% recall, and a 99.74% F1-score, significantly outperforming existing methods such as deep learning, bidirectional transformers, and graph convolutional networks. By combining high accuracy with enhanced interpretability, AC-ITCT provides a robust and efficient solution for real-time arrhythmia detection in clinical settings.</p>