Feature-based ML models reduce next-week forecasting MAE by 22-33% over heuristic baselines for weekly minutes practiced and new skills mastered¶

Claim · All claims
Evidence · 1 study · associational r2 · q2
Currency · newest study 2026 · checked against its sources 2026-10-20 (automated, 1 of 1 entry) · not reviewed by a person · recheck by 2029: tool and model accuracy changes quickly

Subclaims¶

q2 i? Supervised feature-based models substantially outperform heuristic baselines on week-ahead engagement forecasting, reducing error by 22% for minutes and 33% for skills. → Eric S. Qiu 2026

Evidence¶

Eric S. Qiu 2026¶

Eric S. Qiu, Danielle R. Thomas, Boyuan Guo, Vincent Aleven, Conrad Borchers. (2026). From Heuristics to Analytics: Forecasting Effort and Progress in Online Learning. https://arxiv.org/abs/2605.12788 doi:10.48550/arxiv.2605.12788

q2 · i? · associational · r2

Benchmark of fifteen predictors (regressions, decision trees, neural networks) on ITS logs from 425 middle-school students, with student-level splits; the article reports that on average, error falls by "22%for minutes and33%for skills" relative to baselines, with bootstrap CIs excluding zero.

"On average, error falls by22%for minutes and33%for skills."

Discussion¶

Related Claims¶

• Ablations show every feature group contributes to skills forecasting, while recent engagement activity alone nearly suffices for minutes — related
• Effort forecasting is driven by recent engagement activity while progress forecasting depends more on learner-centered signals such as ability, difficulty, and consistency — related
• Percentile heuristics transferred from health behavior systematically overpredict weekly K-12 engagement and underperform XGBoost by 20-30% — related
• TRACE outperforms LSTM, GNN and XGBoost baselines on grade prediction — related