In this paper, we present a framework to apply deep learning techniques to classify pre-oral cancer cells into four categories: normal, mild, moderate, severe. We use state-of-the-art classification techniques to analyze microscopic images and accurately distinguish between normal, mild, moderate, severe categories. Four classification models are trained on all three magnifications (having four classes each) individually, and it is observed that the 10 \(\times \) magnification yields improved performance compared to the other magnification. Considering 10 \(\times \) magnification as the optimal magnification for pre-oral cancer detection, we extend our study to incorporate stain normalization, hierarchical classification, and ensemble modeling to uncover intricate dependencies within the feature space and enhance classification accuracy. Based on the experiments conducted on the custom dataset, we conclude that selecting the optimal magnification (10 \(\times \) ) is necessary but not sufficient to improve the accuracy of pre-oral cancer classification. This work finds its application in histopathological image analysis which plays an important in diagnosing oral cancer. With advances in computer vision techniques, automated systems can assist in classifying histopathological images, providing valuable support to pathologists. These systems help improve diagnostic accuracy and efficiency, ultimately contributing to improved oral cancer detection and treatment.