E-Learning Performance Prediction: Mining the Feature Space of Effective Learning Behavior.

Area of Science:

• Educational Technology
• Data Science
• Learning Analytics

Background:

• Online education relies heavily on learning analytics to predict student performance and provide timely feedback.
• Identifying key learning behaviors and their relationship with academic outcomes is crucial for effective prediction models.
• Understanding the impact of learning behavior classification on prediction accuracy is an ongoing research challenge.

Purpose of the Study:

• To propose a self-adaptive feature fusion strategy to enhance the performance of learning performance prediction models.
• To mine an effective e-learning behavior feature space by classifying learning behaviors.
• To improve the accuracy and reliability of predicting learners' academic success or failure in online environments.

Main Methods:

• Developed an E-learning Behavior Classification (EBC) Model based on interaction objects and learning processes.
• Applied entropy weight and variance filtering methods for preliminary feature space reduction.
• Integrated the EBC Model with a self-adaptive feature fusion strategy to construct a learning performance predictor.

Main Results:

• Identified basic interactive behavior (BI) and knowledge interaction behavior (KI) as having the strongest correlation with learning performance.
• Demonstrated that the self-adaptive feature fusion strategy significantly improves predictor performance.
• Achieved high performance metrics: 98.44% accuracy (ACC), 0.9893 F1-score (F1), and 0.9600 kappa (K).

Conclusions:

• The proposed strategy effectively mines valuable features for learning performance prediction.
• The study provides a novel perspective on constructing e-learning performance predictors.
• Findings offer valuable insights for online learners and educational managers to improve e-learning experiences and outcomes.