TRACE outperforms LSTM, GNN and XGBoost baselines on grade prediction¶

Claim · All claims
Evidence · 1 study · design r2 · q2
Currency · newest study 2026 · not yet checked against its sources · not reviewed by a person · recheck by 2029: tool and model accuracy changes quickly

Subclaims¶

q2 i? TRACE achieved MSE approximately 30% lower and MAE approximately 15% lower than two LSTM models, MSE/MAE 15%-20% lower than a GNN, and drastically outperformed XGBoost. → Paul Savala 2026

Evidence¶

Paul Savala 2026¶

Paul Savala. (2026). Jointly Predicting Courses and Grades Using a Transformer-Based Model. arXiv. https://arxiv.org/abs/2608.13409 doi:10.48550/arxiv.2608.13409

q2 · i? · design · r2

Comparison experiment on the same institutional dataset (Table 1: TRACE MSE 0.0392/MAE 0.1339; EncDecLSTM 0.0530/0.1608; UniLSTM 0.0544/0.1619; GNN 0.0495/0.1640; XGBoost 1.3317/0.8818). The text reports "a MSE approximately 30% lower than the LSTM models".

"Compared to the two LSTM models, the Transformer model shows a significant improvement, with a MSE approximately 30% lower than the LSTM models, and a MAE approximately 15% lower."

Discussion¶

Related Claims¶

• Jointly predicting courses and grades reduces grade-prediction MAE by 46.4% versus an identical Transformer that predicts grades only — related
• KL-divergence and binary cross-entropy losses for course prediction yield grade prediction within 1-2% of each other — related
• Removing student major from the input marginally improves TRACE performance, suggesting major is recoverable from course history — related
• LSTM and LSTM+SK outperform the Bayesian models on predicting quantized learning gains — related
• Feature-based ML models reduce next-week forecasting MAE by 22-33% over heuristic baselines for weekly minutes practiced and new skills mastered — related
• The DMO-GRU model achieves the best classification results among the compared models, with 91.2% accuracy, 90.8% F1-score, 90.1% precision and 92.5% recall — related
• DMO-GRU attains the lowest MAE and RMSE and the highest R2 score among the compared regression models for forecasting student success — related
• On the ASSISTments real-world benchmark, the proposed method achieves the best performance across all metrics, significantly outperforming all baselines — related
• Replacing the transformer with a single-layer 2048-unit LSTM drops the average score by 5.6, with the LSTM winning only on MRPC — related
• Course prediction achieves modest classification metrics (F1 0.4432) yet the auxiliary task regularizes representations and improves grade prediction — related