On one graphical differentiation item, eye tracking data predicted students' specific difficulties as a multiclass classification problem¶

Claim · All claims
Evidence · 1 study · associational r2 · q2
Currency · newest study 2025 · checked against its sources 2026-10-27 (automated, 1 of 1 entry) · not reviewed by a person

Subclaims¶

q2 i? Three machine learning algorithms (KNN, Random Forest, SVM) applied to eye tracking features achieve an MCC around 0.78 in multiclass classification of five solution categories (correct, slope-height error, block-slope error, inverted-slope error, guess) on the one graphical differentiation item analysed; the authors propose, but did not test, transfer to other items. → Kennel 2025

Evidence¶

Kennel 2025¶

Kennel, K., Ishimaru, S., Küchemann, S., Steinert, S., Kuhn, J., & Ruzika, S. (2025). Gaze-Based Prediction of Students' Math Difficulties - A Time Dynamic Machine Learning Approach to Enable Early Individual Assistance. International Journal of Artificial Intelligence in Education. https://doi.org/10.1007/s40593-024-00447-5

q2 · i? · associational · r2

Machine learning classification study with 143 upper-secondary students using a stationary eye tracker, analysing one graphical differentiation item in depth; features included fixation counts, relative fixation counts, transitions, and saccade direction. The article reports "all 3 classifiers achieve values of the MCC around 0.78" for the multiclass case.

"Overall, all 3 classifiers achieve values of the MCC around 0.78, whereby the values after 50 s are already very good with values above 0.7 (for the KNN algorithm even a little earlier)."

Discussion¶

Related Claims¶

• Binary classification detects slope-height and block-slope difficulties earlier than other difficulties — related
• Random forest algorithms yielded the best classification performance in K–8 MMLA studies comparing multiple machine learning models — related
• Machine learning methods outperformed logistic regression in classification accuracy across reviewed studies — a broader claim this one bears on
• Machine learning models classify the three learning stages from EEG features with over 80% test accuracy, SVM highest at 82.6% — related
• A support vector machine classifier outperformed decision tree and random forest models in predicting cognitive engagement levels of online discussion posts — related
• An LSTM-based strategy temporal analytics framework outperforms random forest, SVM, and majority-class baselines for stealth assessment predictive accuracy — related
• A stacking ensemble of LG, SVM, NB, KNN and XGB achieved the best fine-tuning F1-score (0.88), outperforming individual models such as SVM (0.86) — related
• All three trained classifiers outperformed the zero-rule baseline (28.4% accuracy) for classifying cognitive engagement in discussion posts — related
• A random forest classified flagged versus non-flagged word problems with the best accuracy of five tested models (AUC = 0.75) — related