Recursive feature elimination mitigates overfitting and improves STEM/non-STEM prediction performance, with LR-DKT&SP performing best after RFE.¶

Claim · All claims
Evidence · 1 study · associational r1 · q2
Currency · newest study 2018 · not yet checked against its sources · not reviewed by a person

Subclaims¶

q2 i? Applying RFE to reduce feature dimensionality improves model performance, and after RFE the LR-DKT&SP model performs best on all evaluation measures (AP 0.458, AUC 0.694, combined score 1.280, RMSE 0.414). → Yeung 2018

Evidence¶

Yeung 2018¶

Yeung, C.-K., Lin, Z., Yang, K., & Yeung, D.-Y. (2018). Incorporating Features Learned by an Enhanced Deep Knowledge Tracing Model for STEM/Non-STEM Job Prediction. arXiv preprint arXiv:1806.03256. https://arxiv.org/abs/1806.03256

q2 · i? · associational · r1

Dimensionality-reduction experiments using PCA, RFE and oversampling (Section 5.1, Table 3). The authors report "the RFE process is the most effective for relieving the overfitting issue", and after RFE, LR-DKT&SP's test AP, AUC and combined score rise from 0.341 to 0.458, 0.582 to 0.694 and 1.141 to 1.280 respectively, with RMSE reduced from 0.442 to 0.414.

"Among all the attempted methods, the RFE process is the most effective for relieving the overfitting issue and improving the model performance."

Discussion¶

Related Claims¶

• Models trained with combined DKT/DKT+ knowledge-state and student-profile features generally outperform models trained with student profile alone on STEM/non-STEM job prediction. — related
• LDA projection visualizations show knowledge-state features separate STEM and non-STEM classes better than student-profile features. — related
• The two-fold training process suffers from overfitting due to the small labeled training set relative to feature dimensionality. — related