Alzheimer’s disease (AD) is the most prevalent form of dementia, affecting millions worldwide through gradual neurodegeneration. Early detection is essential for timely intervention, yet conventional diagnostic methods, including neuropsychological testing and imaging, often struggle with early-stage AD identification. This study presents a novel convolutional neural network (CNN) model designed to classify AD stages based on magnetic resonance imaging (MRI) data. The model has achieved a high accuracy of 99.51% so far, with further ablation studies going on. MRI images from Kaggle were preprocessed through normalization, resizing to 128 × 128 pixels, and data augmentation to support reliable classification across four AD stages: Non-demented, very mild demented, mild demented, and moderate demented. A detailed ablation study has improved the CNN’s architecture through iterative adjustments of layers, hyperparameters, and pooling techniques. Furthermore, the model’s generalizability was confirmed on an additional dataset, where it maintained an accuracy of 99.39%. These findings highlight the model’s superior performance and computational efficiency, suggesting its potential as an effective clinical tool. This research contributes to advancing Alzheimer’s disease detection via deep learning, providing a diagnostic tool that could improve clinical decision-making and patient care outcomes, while also floating an idea that because of the extensive data augmentation techniques were used to enhance model robustness, reducing the need for explicit regularization methods such as dropout.