Abstract:

Universities worldwide face persistent challenges related to student dropout, which carries significant academic, economic, and social costs. Although prior research has identified numerous contributing factors, accurately predicting which students are at risk remains a challenge. This study builds and evaluates predictive models designed to recognise students prone to dropping out. Using a large-scale longitudinal dataset containing demographic characteristics, academic performance indicators, enrolment patterns, and behavioural metrics, we develop logistic regression and random forest models to estimate dropout probability and identify the most influential predictors.

The results show that predictive analytics can effectively support proactive student success strategies by enabling early identification of at-risk students. Such insights can help institutions target advising, tutoring, or financial support more systematically. This work contributes to the educational data mining literature by providing an empirical evaluation of machine learning models for dropout prediction and discussing the practical challenges of integrating predictive tools into academic decision-making. Future work will explore real-time predictive systems and assess the impact of targeted interventions informed by these models.

Keywords:

Student dropout, higher education, predictive modeling, machine learning, educational data mining.