Quick prevention and efficient treatment for Alzheimer’s disease (AD) rely upon early diagnosis. This study proposes a unique method that combines deep learning strategies with the Synthetic Minority Oversampling Technique (SMOTE) to overcome the inherent class imbalance in an AD dataset. Our proposed convolutional neural network (CNN) model is designed to accurately detect AD in its early stages, thus offering significant potential to enhance diagnostic outcomes. Through the integration of SMOTE, we ensure a balanced representation of all classes, improving the model’s robustness. Extensive experiments and detailed analysis validate the effectiveness of this approach, demonstrating its superior performance in comparison to traditional methods. The promising results indicate the potential of our method to contribute to early AD detection and improve patient care.