A Factorization Deep Product Neural Network for Student Physical Performance Prediction.

Area of Science:

• Educational Technology
• Sports Science
• Data Science

Background:

• Student physical education (PE) is vital but often overlooked due to academic pressure.
• Accurate prediction of PE performance is essential for targeted interventions and student development.
• Existing performance prediction methods lack comprehensive feature interaction analysis.

Purpose of the Study:

• To propose and evaluate a novel method for predicting physical education course scores.
• To assess the effectiveness of the proposed method against traditional and deep learning approaches.
• To investigate the impact of feature interactions on prediction accuracy.

Main Methods:

• Development of a factorization deep product neural network (FDPNN) for PE score prediction.
• Comparison with established methods: Logistic Regression (LR), Support Vector Machines (SVM), Factorization Machines (FM), and Deep Neural Networks (DNN).
• Utilized Principal Component Analysis (PCA) for dimensionality reduction and data preprocessing.

Main Results:

• The FDPNN demonstrated superior prediction performance on the sports education dataset compared to LR, SVM, FM, and DNN.
• Significant improvements in accuracy, Area Under the Curve (AUC), recall, and F1-score were observed.
• The FDPNN effectively learned complex feature interactions (first, second, and high-order), outperforming single-feature learning.

Conclusions:

• The proposed FDPNN offers a robust and accurate approach for predicting student physical education performance.
• Automated feature interaction learning enhances prediction accuracy and provides deeper insights into performance determinants.
• PCA proved effective for data processing, outperforming benchmark models in dimensionality reduction.