Oral cancer is a major risk to the global health and leads to high death rates and significant suffering. Early detection is vital to improving the outcome of treatment. This study uses deep learning techniques, parallel neural networks (convolutional neural networks, CNNs), and transfer learning methods to propose reliable machine learning systems for oral cancer detection. The proposed model combines CNN and transfer learning to classify patients with oral cancer. Although the dataset used was small and the dispersed class distribution was large, the model was trained and evaluated using reliable metrics such as accuracy, recall, F1 score, and area under the receiver operating characteristic (ROC) curve (AUC). AUC-ROC analysis and confusion matrices were used to verify these results and ensure their reliability. The model obtained an F1 score of 81.48%, an accuracy of 84.62%, with a recall of 78.57%, and an ROC-AUC score of 0.9082 in the test dataset. During training, it achieved 96.94% accuracy, 97.92% precision, 96.17% recall, 97.04% F1 score, and 0.9967 ROC-AUC score. This research highlights how artificial intelligence can affect clinical workflows in early cancer detection. The results provide a hopeful path for the development of automated cancer diagnosis technology.