Construction of a multi-dimensional predictive model for college students' academic performance based on deep...

Area of Science:

• Educational Data Mining
• Machine Learning in Education
• Student Performance Prediction

Background:

• Traditional models struggle with temporal and behavioral student data.
• Deep learning offers enhanced accuracy and adaptability for predictive analysis.
• Accurate academic performance prediction is vital for identifying at-risk students.

Purpose of the Study:

• To develop a robust predictive model for student academic performance.
• To integrate temporal, behavioral, and demographic features for enhanced prediction.
• To leverage multi-dimensional student data for advanced educational analytics.

Main Methods:

• Data collection from diverse sources (grades, LMS, surveys, demographics) for 2000 students.
• Data preprocessing including KNN Imputation, outlier removal, normalization, and PCA for dimensionality reduction.
• Development and optimization of a novel GateLSTMU-Dove model for capturing temporal dependencies and optimizing parameters.

Main Results:

• The GateLSTMU-Dove model achieved a classification accuracy of 98.85%.
• Demonstrated superior performance with lower error metrics compared to baseline methods.
• Provided accurate forecasting and interpretable temporal patterns in student performance.

Conclusions:

• The GateLSTMU-Dove model effectively predicts academic outcomes using multi-dimensional student data.
• Offers interpretable insights to support early intervention strategies.
• Presents a scalable and reproducible approach for data-driven academic performance management.