Explainable machine learning for sustainable education: Predicting college students' reliance on generative...

Area of Science:

• Educational Technology
• Artificial Intelligence in Education
• Human-Computer Interaction

Background:

• Generative artificial intelligence (GenAI) is increasingly integrated into educational settings.
• Students utilize GenAI for knowledge acquisition, problem-solving, and creative exploration.
• Concerns exist regarding potential overreliance on GenAI tools in academic contexts.

Purpose of the Study:

• To assess and manage student dependence on GenAI for sustainable educational use.
• To develop and validate machine learning models for predicting GenAI dependence levels.
• To identify key factors influencing student GenAI dependence.

Main Methods:

• Collected questionnaire data from university students in China.
• Applied principal component analysis and numerical binning for data preprocessing.
• Trained and compared six machine learning models, with Random Forest (RF) showing the best performance (F1-score 0.836).
• Interpreted RF model using SHAP and PDP methods for explainability.

Main Results:

• Classroom speaking pressure was identified as the primary predictor of GenAI dependence, explaining 22.9% of the variance.
• Higher speaking pressure correlated with increased GenAI dependence, particularly in highly dependent groups.
• Model robustness was confirmed through ablation studies and multi-dimensional sensitivity analyses.

Conclusions:

• Proposed an effective and explainable method for predicting GenAI dependence.
• Provided empirical evidence to guide students toward responsible GenAI usage.
• Informed targeted interventions to balance AI support with autonomous learning in education, aligning with UN SDG 4.