Predicting performance of students by optimizing tree components of random forest using genetic algorithm.

Area of Science:

• Educational Data Mining
• Machine Learning Applications
• Academic Performance Prediction

Background:

• Student academic performance prediction faces challenges with existing methods due to poor generalizability and interpretability.
• Current prediction models often lack the reliability needed for effective educational interventions.

Purpose of the Study:

• To develop a more accurate and reliable approach for predicting student academic performance.
• To address the limitations of low generalizability and lack of interpretability in existing prediction methods.

Main Methods:

• A hybrid feature selection scheme combining Information Gain (IG) and Laplacian Score (LS) for ranking.
• Sequential Forward Selection (SFS) to identify the most relevant indicators.
• A multi-class classification model integrating Random Forest with a Genetic Algorithm (GA).

Main Results:

• The proposed approach achieved an average prediction accuracy of 93.11% in a case study.
• Demonstrated a minimum improvement of 2.25% compared to baseline prediction methods.
• Validation through Accuracy, Precision, Recall, and F-Measure metrics confirmed the model's efficiency.

Conclusions:

• The combined IG-LS feature selection and RF-GA classification offers a robust and interpretable solution for student performance prediction.
• The developed strategy significantly enhances prediction accuracy and reliability over traditional methods.
• This approach provides a valuable tool for educators and institutions to identify students at risk and implement timely support.