Jointly Predicting Courses and Grades Using a Transformer-Based Model
TRACE cuts grade-prediction error by nearly half by modeling next semester’s courses and grades together.
The paper says course load matters because courses happen concurrently within a semester, not as a flat sequence. TRACE encodes academic history by semester, predicts the upcoming course set, and predicts grades for those courses with a combined loss. Trained on ten years of institutional data, it beat a grades-only version of the same architecture, LSTM sequence models, and graph neural network approaches. The author frames it as a possible component for early detection systems in higher education. ArXiv · AI/CL/LG's note
The paper says course load matters because courses happen concurrently within a semester, not as a flat sequence. TRACE encodes academic history by semester, predicts the upcoming course set, and predicts grades for those courses with a combined loss. Trained on ten years of institutional data, it beat a grades-only version of the same architecture, LSTM sequence models, and graph neural network approaches. The author frames it as a possible component for early detection systems in higher education. ArXiv · AI/CL/LG's note
score 4