Optimized ensemble deep learning for predictive analysis of student achievement

Area of Science:

• Educational Technology
• Artificial Intelligence in Education
• Data Mining

Background:

• Education is crucial for personal and societal advancement.
• Technological advancements, particularly AI, are transforming learning accessibility and methods.
• Higher education integrates technology to improve traditional teaching.

Purpose of the Study:

• To present an innovative hybrid approach for predicting student performance in educational settings.
• To address the challenges of increasing data volume in graduate and postgraduate programs.
• To improve the efficiency and accuracy of educational data mining.

Main Methods:

• A hybrid model combining DistilBERT with Long Short-Term Memory (DBTM) was developed.
• The Spotted Hyena Optimizer (SHO) was employed to optimize the parameters of the DBTM model.
• The proposed DBTM-SHO approach was evaluated on extensive datasets.

Main Results:

• The DBTM-SHO model demonstrated significant improvements in accuracy, log loss, and execution time compared to previous models.
• Achieved 98.7% accuracy and 0.03% log loss.
• Reduced processing time by 15-25% through optimization, effectively handling large datasets.

Conclusions:

• The DBTM-SHO approach offers a robust solution for student performance prediction in the era of big data.
• This method represents a significant advancement in educational data mining.
• Provides a strong foundation for institutions evaluating student achievement with large datasets.