Alzheimer’s disease is a devitalizing condition which leads to cognitive decline, and identifying it early is essential for timely treatment and improving the quality of life for patients. Medical imaging plays a key role in diagnosing Alzheimer’s disease; however, automated tools are needed for accurate and efficient classification. A novel approach, SMoViT-AD, is proposed to predict Alzheimer’s disease from medical images by integrating the MobileViT architecture with SMOTE-Tomek for dataset balancing. The model achieved exceptional performance, with accuracy, precision, recall, and F1-score of 99.67%, significantly outstripping state-of-the-art models employed. SMoViT-AD not only provides high accuracy but also enhances the transparency of its predictions, by incorporating Grad-CAM for model interpretability. The outcomes establishes the efficacy of the proposed model in improving diagnostic correctness, making it a powerful tool for early Alzheimer’s disease detection. Informed decision-making by healthcare personnel could be facilitated by the high accuracy and interpretability, resulting in improved patient outcomes.