Discriminable Multi-Label Attribute Selection for Pre-Course Student Performance Prediction.

Area of Science:

• Educational Technology
• Machine Learning
• Student Support Systems

Background:

• Traditional university teaching faces challenges with low teacher-student ratios hindering personalized learning follow-up.
• Existing early warning systems often rely on e-learning data or predict risks late in the course.
• Prior research has overlooked feature redundancy in student learning data for predictive modeling.

Purpose of the Study:

• To develop a pre-class student performance prediction model for traditional classroom settings.
• To identify high-risk students before a course commences, enabling timely intervention.
• To address limitations of existing learning early warning systems.

Main Methods:

• Transformed pre-class student performance prediction into a multi-label learning problem.
• Applied attribute reduction to streamline characteristic information from previously learned courses.
• Explored relationships between prior course characteristics and attributes of courses to be taken.

Main Results:

• The proposed multi-label learning approach demonstrated superior performance on 10 real-world datasets.
• Achieved better results in multi-label classification evaluation metrics compared to advanced methods.
• Successfully identified high-risk students prior to course commencement.

Conclusions:

• The developed model offers an effective solution for early risk detection in traditional academic environments.
• Attribute reduction enhances the efficiency and accuracy of student performance prediction.
• This approach provides a valuable tool for proactive student support and academic success.