Predicting the Performance of Students Using Deep Ensemble Learning.

Area of Science:

• Educational Technology
• Artificial Intelligence
• Machine Learning

Background:

• Accurate student performance forecasting is crucial for universities to improve academic outcomes and reduce attrition.
• Technology-enhanced learning generates large datasets offering insights into student knowledge and engagement.
• Analyzing these datasets facilitates data-driven educational strategies.

Purpose of the Study:

• To develop an accurate student academic performance prediction model.
• To introduce a novel feature-ranking mechanism for identifying key performance indicators.
• To optimize the training and configuration of deep neural network ensembles.

Main Methods:

• An ensemble of deep neural networks was utilized for academic achievement forecasting.
• A new feature-ranking mechanism was developed to identify pertinent student performance predictors.
• An optimization strategy was employed for concurrent configuration and training of the deep neural networks.
• Weighted voting was implemented within the ensemble for enhanced prediction accuracy.

Main Results:

• The proposed method achieved a Root-Mean-Square Error (RMSE) of 1.66, a Mean Absolute Percentage Error (MAPE) of 9.75, and an R-squared value of 0.7430.
• These results significantly outperformed a null model (RMSE = 4.05, MAPE = 24.89, R-squared = 0.2897).
• The findings demonstrate the effectiveness of the ensemble and optimization techniques.

Conclusions:

• The developed ensemble model with optimized deep learning parameters accurately predicts student academic performance.
• The novel feature-ranking mechanism effectively identifies crucial factors influencing student success.
• The proposed approach offers a significant advancement in educational data analytics for student support.