Alzheimer’s disease (AD) represents one of the most debilitating neurodegenerative disorder that progressively affects memory, cognition, and the ability to perform routine activities. While early diagnosis is vital, but challenging with traditional methods that rely on clinical assessments and neuroimaging analysis. This research analyzes a deep learning-based approach using ResNet50, VGG16, and EfficientNet-B4 for automated AD classification into four stages: Mild Cognitive Impairment (MCI), Mild AD, Moderate AD, and Severe AD. Transfer learning is applied to enhance feature extraction, while data preprocessing and augmentation are used to address class imbalance. Model training and evaluation were conducted using a benchmark dataset with a train-test-validation split of 70:15:15. Among the models, ResNet50 achieved the optimal result of 95 % and an AUC-ROC of 0.98, outperforming traditional approaches. The results highlight the capability of deep learning in early and automated AD detection, offering a reliable tool for clinical decision-making.