Timely detection of Oral Squamous Cell Carcinoma (OSCC) is pivotal for enhancing treatment efficacy. Current diagnostic paradigms, however, fall short in early detection due to limited data diversity and volume. Our study bridges this gap by introducing a novel machine learning (ML) framework that incorporates Generative Adversarial Networks (GANs) and Variational Autoencoders (VAEs) for synthetic data augmentation, enriching the dataset beyond the reach of traditional diagnostic methods. This research departs from existing works which predominantly rely on direct ML application with conventional datasets, by employing these sophisticated techniques to generate high-fidelity, diverse synthetic datasets that closely mimic rare positive OSCC instances. This targeted augmentation effectively counters the data scarcity and model overfitting issues prevalent in current literature, ensuring robust model training and validation. We devised a robust methodology that harnesses the synergistic potential of GANs and VAEs with an ensemble of advanced ML algorithms, including CatBoost, XGBoost, LightGBM, Random Forest, and Deep Neural Networks. The outcomes of our approach are significant, with our models demonstrating enhanced predictive accuracy, validated by rigorous testing against established benchmarks. The findings indicate that synthetic data augmentation via GANs and VAEs can revolutionize OSCC screening, offering a substantial improvement over conventional methods and contributing to a paradigm shift in early cancer detection strategies.