A PREDICTIVE MODEL FOR STUDENT RETENTION PATTERNS AND HOLISTIC RISK ASSESSMENT IN HIGHER EDUCATION SYSTEM

Abstract

Student retention remains a critical concern in higher education, with significant implications for institutional effectiveness, student success, and resource allocation. This study presents a predictive modelling approach to analyze students’ retention patterns and support holistic student risk assessment. Using historical academic, demographic, and behavioural data, the research develops and evaluates machine learning models to identify students at risk of attrition. Key features such as academic performance, attendance records, socio-economic background, and engagement indicators are incorporated to capture a comprehensive view of student progression. Using a predictive modelling approach, a balanced Random Forest classifier was implemented to categorize students into Low, Medium, and High-risk groups. To address the interpretability limitations of machine learning models, SHAP (SHapley Additive exPlanations) was employed to provide transparent and explainable insights into the model's predictions. The resulting outputs show that academic performance, in particular, Grade Point Average (GPA), and attendance rate constitute the most significant predictors of student risk. The result further reveals that behavioural engagement metrics, including LMS login frequency, session duration, and forum participation, provide complementary insights, which enable earlier detection of disengagement patterns. The model, as shown by the confusion matrix, achieves a precision of 92%, demonstrating its effectiveness in capturing complex, non-linear relationships among variables.