Neural network-based models have found wide use in automatic long-term electrocardiogram (ECG) analysis. However, such black box models are inadequate for analysing physiological signals where credibility and interpretability are crucial. Indeed, how to make ECG analysis transparent is still an open problem. In this study, we develop a Tsetlin machine (TM) based architecture for premature ventricular contraction (PVC) identification by analysing long-term ECG signals. The architecture is transparent by describing patterns directly with logical AND rules. To validate the accuracy of our approach, we compare the TM performance with those of convolutional neural networks (CNNs). Our numerical results demonstrate that TM provides comparable performance with CNNs on the MIT-BIH database. To validate interpretability, we provide explanatory diagrams that show how TM makes the PVC identification from confirming and invalidating patterns. We argue that these are compatible with medical knowledge so that they can be readily understood and verified by a medical doctor. Accordingly, we believe this study paves the way for machine learning (ML) for ECG analysis in clinical practice.