This study critically examines the application of machine learning technologies in higher education and academic management, with particular emphasis on their use as early warning and intervention systems for identifying students at risk of academic dropout. Employing a case study approach, the research focuses on a Chinese public-sector university recognised as a pioneer in the adoption of digital learning technologies and learning management systems (LMS). The empirical analysis is based on observational data and academic records of 141 undergraduate (BS) students enrolled across multiple programmes and disciplines. The results indicate that approximately 17% of the analysed students were classified as at risk of dropout. Risk identification was conducted using machine learning models that integrated traditional academic indicators, such as cumulative GPA and current GPA, alongside behavioural and engagement-related variables. Key behavioural factors included assignment submission timeliness, attendance rates, LMS login frequency, and completed credit hours, with higher values in these indicators corresponding to a lower likelihood of dropout. Additionally, student engagement, for which online forum participation benchmark has been adopted in this study, reflects a better assessment of levels of academic engagement, as students exhibiting higher levels of online interaction with peers and instructors demonstrated a reduced risk of attrition. Among the various machine learning models evaluated, XGBoost emerged as the most effective in predicting at-risk students, achieving superior accuracy and precision compared to alternative approaches.
