This study introduces a hybrid predictive architecture grounded in the neutrosophic framework, aimed at enhancing the accuracy of student learning outcome predictions. The proposed approach addresses challenges related to data uncertainty, indeterminacy, and noise by embedding deep learning models, specifically CTGAN and Transformer, within a neutrosophic encoder-decoder structure. At the core of the architecture, neutrosophic logic functions as the key mechanism for managing ambiguous and incomplete information, which is commonly encountered in educational datasets. Experimental evaluations on real-world data, incorporating a targeted noise-injection strategy, demonstrate that the proposed neutrosophy-based model consistently outperforms traditional deterministic methods. The model achieves a minimum MSE of 0.018 and a maximum R2 of 96.05%, representing a significant improvement over previous state-of-the-art predictive performances.