Students learning performance prediction based on feature extraction algorithm and attention-based bidirectional...

Area of Science:

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

Background:

• Student grade prediction is crucial for educational improvement, but current models struggle with noisy public datasets.
• Weakly correlated factors in educational data often negatively impact predictive model performance.
• There is a need for robust models to optimize teaching, learning, and parental guidance.

Purpose of the Study:

• To develop and identify the optimal data mining model for accurate student grade prediction.
• To address the limitations of existing models in handling educational datasets.
• To provide data-driven policy recommendations for educational modernization.

Main Methods:

• Feature extraction and dimensionality reduction using Factor Analysis (FA).
• Grade prediction employing a Bidirectional Gate Recurrent Unit (BiGRU) model integrated with an attention mechanism.
• Comparative analysis against single models like Linear Regression (LR), Back Propagation (BP), Random Forest (RF), and Gate Recurrent Unit (GRU).

Main Results:

• The proposed FA-BiGRU-attention model demonstrated superior prediction accuracy compared to all benchmark models.
• The model exhibited consistent performance across various multi-step prediction scenarios.
• Ablation experiments validated the effectiveness of the combined FA, BiGRU, and attention components.

Conclusions:

• The FA-BiGRU-attention model offers a significant advancement in predicting student academic performance.
• This approach enables proactive identification of learning difficulties and influencing factors.
• The findings support the transformation of educational practices and talent development through data-driven insights.