Predicting Student Dropout in Higher Education: An Ensemble Learning Approach with Feature Importance Analysis

Student dropout in higher education remains a global challenge, particularly in developing regions where early interventions are hindered by reliance on traditional indicators like GPA or attendance. This study addresses the issue by proposing a predictive model using ensemble machine learning techniques, integrating Logistic Regression, Random Forest, and AdaBoost. These models were combined using soft and hard voting classifiers to enhance prediction accuracy and reliability. The dataset, comprising 4,424 student records, includes demographic, academic, and socio-economic features. Results showed that the soft voting ensemble achieved the highest accuracy (80.56%) and AUC (91%), outperforming individual classifiers. Feature importance analysis revealed academic performance, tuition status, and parental background as key predictors of dropout. The model not only identifies at-risk students with high precision but also offers actionable insights for early intervention. This approach equips higher learning institutions with data-driven strategies to improve retention and student success outcomes.