Deep learning-based AI model for predicting academic success and engagement among physical higher education students

Area of Science:

• Educational Technology
• Machine Learning in Education
• Sports Science

Background:

• Academic achievement and student engagement are critical in physical education (PE).
• Predicting these outcomes is challenging due to the intersection of cognitive and physical skills.
• Early identification of students at risk is crucial for timely intervention.

Purpose of the Study:

• To develop and evaluate a stacked ensemble machine learning model, HybridStackNet, for jointly predicting academic success and engagement in higher education PE students.
• To assess the performance of HybridStackNet against baseline models.
• To explore the explainability of the model's predictions using techniques like Partial Dependence Plots (PDPs) and LIME.

Main Methods:

• A stacked ensemble model (HybridStackNet) was designed using Random Forest and Support Vector Machine (SVM) as base learners and Logistic Regression as a meta-learner.
• Data preprocessing involved label encoding, z-score normalization, feature selection, and SMOTE for class balancing.
• Stratified 5-fold cross-validation and GridSearchCV were employed for model evaluation and hyperparameter tuning.

Main Results:

• HybridStackNet achieved high performance metrics, including Accuracy (0.992), Precision (0.9922), Recall (0.992), and F1-score (0.9915).
• The model significantly outperformed several baseline machine learning models.
• Explainability methods identified key features like Attendance Rate and Motivation Level influencing predictions.

Conclusions:

• HybridStackNet offers a highly accurate and interpretable machine learning solution for predicting student success in PE.
• The model facilitates early detection of performance risks, enabling targeted support.
• This approach enhances educational strategies in PE by providing data-driven insights.