The World Health Organization's 2022 report states that oral cancer affects 3.5 billion people worldwide and is responsible for an expected 177,757 fatalities in 2020. Oral cancer is hazardous because it is difficult to detect in its early stages. To solve this problem, our research introduces the ground-breaking deep learning framework OCS-Net. The framework's goal is to categorize the different stages of oral cancer utilizing CT scan pictures from S.M.S. Hospital in Jaipur patients. We ran into the issue of class imbalance in the dataset during the data collection procedure. To solve this problem, we used generative adversarial networks to create artificial images that represented all classes fairly. Additionally, we modified the convolutional neural network parameters used in OCS-Net using genetic algorithm methods. The OCS-Net evaluation produced remarkable results with an AUC-Score of 98.07% and an average accuracy of 98.05%. In terms of performance and accuracy, OCS-Net outperformed well-known transfer learning models including VGG-16, VGG-19, ResNet-101, and MobileNet. This study underlines the value of applying deep learning approaches to medical image processing while also showcasing the potential of OCS-Net as a useful tool for precise oral cancer stage evaluation.