Modeling student satisfaction in online learning using random forest.

Area of Science:

• Educational Technology
• Psychology
• Data Science

Background:

• Online education is rapidly expanding, increasing the need to understand student satisfaction with digital platforms.
• Previous research often isolated technical or pedagogical factors, neglecting their interaction with psychological well-being through nonlinear mechanisms.

Purpose of the Study:

• To investigate the multifactorial determinants of university students' satisfaction with digital learning platforms.
• To model satisfaction using a machine learning framework that accounts for nonlinear interactions between technical, pedagogical, and psychological factors.

Main Methods:

• A Random Forest framework was used to model satisfaction with data from 782 university students.
• Variables included platform usability, content quality, emotional experience, and self-regulation.
• Data preprocessing involved Z-score standardization and Synthetic Minority Over-sampling Technique (SMOTE) for class imbalance.

Main Results:

• Platform stability and content update frequency were the most influential predictors of satisfaction (AUC > 0.95).
• Psychological factors, including perceived enjoyment and emotional stability, also significantly contributed to satisfaction.
• Partial dependence plots revealed complex nonlinear patterns, such as threshold and saturation effects, missed by traditional linear models.

Conclusions:

• Machine learning effectively models nonlinear interactions in digital learning satisfaction, integrating cognitive-affective dimensions.
• Actionable insights for platform optimization are provided, emphasizing the importance of both technical stability and psychological engagement.
• Future research should explore additional psychological constructs and diverse populations to enhance model generalizability for inclusive digital education.