Artificial Intelligence-Driven Early Prediction of Student Dropout and Academic Outcomes in Higher Education: A Comparative Study of Advanced Machine Learning Approaches

Student dropout in higher education remains a critical challenge with significant academic, social, and economic implications. Early identification of students at risk of dropout enables institutions to design timely and targeted interventions that support academic success and improve retention rates. This study proposes a machine learning (ML)–driven framework for the early prediction of student dropout and academic outcomes in higher education using a comprehensive, real-world dataset collected from a higher education institution. The prediction task is formulated as a multiclass classification problem with three outcomes: dropout, enrolled, and graduate. To evaluate the effectiveness of different modeling approaches, we conduct a comparative analysis of widely used ML algorithms, including Logistic Regression, Naïve Bayes, k-Nearest Neighbors, Support Vector Machine, Decision Trees, Random Forest (RF), AdaBoost, XGBoost, LightGBM, and CatBoost. Results indicate that ensemble models achieve the best performance. RF attains the highest test accuracy (0.7797) and ROC-AUC (OvR) (0.8919), while LightGBM yields the best Macro-F1 (0.7082). Feature importance analysis shows that early academic progress indicators (approved units and semester grades) are the strongest predictors, followed by selected administrative/contextual factors such as tuition-fee status and course. Overall, this study provides empirical evidence supporting the use of ML techniques as effective decision-support tools for higher education institutions. The proposed framework offers actionable insights for administrators and policymakers seeking to develop data-driven strategies aimed at reducing dropout rates, improving academic success, and promoting equitable access to educational opportunities.