Alzheimer’s disease, a leading cause of dementia, presents substantial diagnostic challenges due to its progressive and multifaceted nature. This paper introduces a novel classification method designed to improve diagnostic accuracy by integrating features from a pre-trained AlexNet model with those obtained from Histogram of Oriented Gradients (HOG). By combining deep learning and traditional image feature extraction techniques and employing mutual information for feature selection, our approach significantly enhances classification performance. We evaluate our method against models using only AlexNet features, only HOG features, and those with feature selection applied to either AlexNet or HOG features. Our results, achieved on a Kaggle dataset, show that our integrated approach achieves a final accuracy of 98.67%, outperforming all compared methods. This demonstrates that our method not only improves diagnostic precision but also reduces computational complexity, providing a robust and efficient framework for Alzheimer’s disease classification.