Optimization vs Boosting: Comparison of Strategies on Educational Datasets to Explore Low-performing at-risk and Dropout Students
Optimization vs Boosting: Comparison of Strategies on Educational Datasets to Explore Low-performing at-risk and Dropout Students

Summary/Abstract: The paper proposes a comprehensive student academic performance prediction approach by integrating machine learning with metaheuristic optimization. Initial models (Logistic Regression, Decision Tree, Random Forest, MLP) were refined using boosting techniques (Gradient Boosting, XGBoost, LightGBM), with XGBoost achieving 95.59% accuracy. Eight modern optimization algorithms were applied for feature selection to enhance model efficiency and interpretability, with the Grey Wolf Optimizer and the Heap-Based Optimizer outperforming others in key metrics. Support Vector Machine algorithms applied after feature selection strengthened the predictive capability of the selected feature subsets. The research outcomes demonstrate that uniting boosting approaches with feature selection algorithms enables the creation of reliable and scalable predictive models that detect student success and failure earlier.

• Details
• Contents