A MACHINE LEARNING APPROACH TO ACADEMIC PERFORMANCEPREDICTION: A COMPARATIVE STUDY OF XGBOOST AND RANDOM FOREST
Keywords:
Performance, Extreme Gradient Boosting, Random Forest, Prediction, Students, Academic Performance.Abstract
Accurately predicting student academic performance is a critical objective in educational data mining and
learning analytics, as it enables data-driven interventions aimed at improving educational outcomes. This
study investigates and compares the predictive performance of two widely used machine learning
algorithms: Random Forest (RF) and Extreme Gradient Boosting (XGBoost) using real-world student data.
A dataset comprising 400 student records collected from six departments at the Federal Polytechnic Ukana
was preprocessed through principal component analysis (PCA) and feature normalization to enhance
model accuracy. Sixteen key features were selected based on their eigenvalues and explained variance. The
dataset was partitioned into training (80%) and testing (20%) subsets. Model performance was assessed
using standard evaluation metrics, including Mean Squared Error (MSE), Mean Absolute Error (MAE), Rsquared (R²), Explained Variance Score (EVS), and Median Absolute Error (MedAE). Experimental results
indicate that both RF and XGBoost exhibit strong predictive capabilities, with RF demonstrating a slight
advantage across several metrics. These findings underscore the potential of machine learning approaches
in educational performance prediction and provide actionable insights for model selection in academic
analytics frameworks
