Learner models with individual-difference features (PEV, student-level proportional correctness) improve cross-validated fit on six educational datasets¶

Claim · All claims
Evidence · 1 study · associational r2 · q2
Currency · newest study 2022 · not yet checked against its sources · not reviewed by a person

Subclaims¶

q2 i? Across six datasets, models including features adjusting for individual differences achieved lower student-stratified cross-validated RMSE than PFAP or RKCPFA. → Eglington 2022

Evidence¶

Eglington 2022¶

Eglington, L.G., Pavlik, P.I. (2022). How to Optimize Student Learning Using Student Models That Adapt Rapidly to Individual Differences. International Journal of Artificial Intelligence in Education. https://doi.org/10.1007/s40593-022-00296-0

q2 · i? · associational · r2

Analysis of six datasets (Andes, Assistments, Cognitive Tutor/KDD, Chinese tones, McGraw Hill, statistics cloze) fit with five logistic-regression model variants using 10-fold student-stratified cross-validation. Improvement was found "in all tested datasets"; no effect size printed.

"Overall, models that included features to adjust for individual differences improved fit relative to PFAP or RKCPFA in terms of student-stratified cross-validated RMSE (see Table 2). We also found a benefit of RKCPFA over PFAP in 4 of 6 datasets."

Discussion¶

Related Claims¶

• The combined model with both PEV and student-level proportional decay typically fit better than either feature alone — a narrower finding that bears on this claim
• Adding PEV or student-level proportional correctness significantly reduces the absolute value of student-level signed error in all six datasets — a narrower finding that bears on this claim
• Adding a Prior Error Valence (PEV) feature to a population-parameter model reduces systematic mastery-timing error in simulation — related
• No single learner model was best across the six datasets, justifying a broad multi-model approach — related