KL-divergence and binary cross-entropy losses for course prediction yield grade prediction within 1-2% of each other¶

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? An ablation comparing the KL-based set loss to binary cross-entropy for course prediction found grade prediction within 1-2% for both MAE and MSE. → 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

Ablation study reported in the training section comparing the novel KL-divergence course-set loss against binary cross-entropy. The article notes "grade prediction within 1-2% of each other" and suggests BCE is a reasonable alternative, while arguing KL has learning-analytics-aligned benefits.

"we ran an ablation study comparing the above loss function to binary cross entry for course prediction, and noted grade prediction within 1-2% of each other between the two models for both MAE and MSE."

Discussion¶

Related Claims¶

• Jointly predicting courses and grades reduces grade-prediction MAE by 46.4% versus an identical Transformer that predicts grades only — related
• TRACE outperforms LSTM, GNN and XGBoost baselines on grade prediction — related
• Fine-tuning LLMs and MLMs with cross-entropy loss using soft targets outperforms standard discretization and regression-head approaches on vocabulary difficulty prediction — related
• Removing student major from the input marginally improves TRACE performance, suggesting major is recoverable from course history — related