Arrhythmias represent a significant category of cardiovascular disorders. Thus, making early diagnosis and accurate classification is essential for effective patient treatment. Electrocardiogram (ECG) signals, which monitor the heart’s electrical activity, serve as the primary method for detecting arrhythmias. In this paper, we present a novel approach to arrhythmia classification by leveraging the image classification capabilities of MobileViT model. Our methodology involves converting ECG signals from the MIT-BIH Arrhythmia Database into scalograms using the Continuous Wavelet Transform (CWT), thereby capturing both time and frequency information. These ECG scalograms are then fed into the MobileViT model for classification into five distinct arrhythmia classes. To enhance the model’s generalization capabilities, we employ the mixup data augmentation technique during the training. Our approach achieved an impressive accuracy of 98.97%, demonstrating the effectiveness of our approach in improving the automated early detection and classification of arrhythmias.