Stage 1 (after the first unit review assignment) is the optimal point for identifying at-risk students, balancing timeliness, accuracy, and fairness¶

Claim · All claims
Evidence · 1 study (2 entries) · design r2 · q2
Currency · newest study 2025 · checked against its sources 2026-10-20 (automated, 2 of 2 entries) · not reviewed by a person

Subclaims¶

q2 i? Models at Stage 1 showed the least bias on average, with all fairness values below the 0.20 threshold, and comparable predictive performance to later stages. → Alison Cheng 2025 q2 i? Early predictions at Stage 1 achieved accuracy similar to predictions using more data at later stages, making it the optimal identification point. → Alison Cheng 2025 (2)

Evidence¶

Alison Cheng 2025¶

Alison Cheng, Bo Pei and Cheng Liu. (2025). Balancing Act: Early, Fair, and Accurate Identification of At-Risk Students. Journal of Learning Analytics, 12(3), 47–65. https://doi.org/10.18608/jla.2025.8761

q2 · i? · design · r2

Analysis of four ML models trained on data from 215 AP Statistics students, evaluated with four fairness metrics across three stages. The article reports "the models at Stage 1 have the least bias, on average" with all values under 0.20.

"Notably, the models at Stage 1 have the least bias, on average, compared to the Demog Stage and Stage 2 , with all values falling below the 0.20 threshold ."

Alison Cheng 2025 (2)¶

Alison Cheng, Bo Pei and Cheng Liu. (2025). Balancing Act: Early, Fair, and Accurate Identification of At-Risk Students. Journal of Learning Analytics, 12(3), 47–65. https://doi.org/10.18608/jla.2025.8761

q2 · i? · design · r2

Comparison of model performance across stages on the imbalanced dataset showed Stage 1 predictions achieved "similar levels of accuracy to those using more data at a later stage", supporting Stage 1 as the optimal identification point.

"In addition, the models at Stage 1 also have a comp arable performance with that of Stage 2 and the Demog Stage, which means that early predictions can achieve similar levels of accuracy to those using more data at a later stage."

Discussion¶

Related Claims¶

• Attendance, behavior, and course performance (the ABC indicators) reliably predict which secondary students are at risk of dropping out — the general claim this finding bears on, with the studies behind it
• Balancing the dataset significantly improved model predictive performance but increased bias under demographic parity — related
• Adding more learning activity data beyond Stage 1 did not significantly improve predictive performance of at-risk identification models — related
• Removing the sensitive race feature caused little impact on predictive performance at Stage 1 and no significant fairness differences, with mixed fairness effects — related
• Post-processing mitigation with ThresholdOptimizer reduced bias on most fairness metrics while maintaining accuracy reasonably well, with a slight performance drop — related
• Preprocessing bias mitigation (DIR, RW, SUP) reduced subgroup disparities in TPR on OULAD, with effectiveness varying by demographic attribute — related
• Incorporating more learning activity data reduced the potential bias caused by overreliance on demographic information — related
• Achievement gap estimates are widely used to measure the effectiveness and fairness of the education system, so their accuracy and unbiasedness are necessary for appropriate conclusions — related
• Stakeholder groups converged on recommendations for timelier results, removal of racial and cultural bias, and more useful, accessible results — related