Survival models for predicting student dropout at university across time

The aim of this study is to develop a tool to recognize major responsible factors of student dropout through time, both in terms of student characteristics and type of degree courses, and to accurately predict the student time to dropout, if any. From a predictive point of view, we aim at developing an early warning system to early predict the status of a student career, identifying the risky timings in terms of dropout, as a supporting tool for early interventions policies. To this end, we follow a Survival Analysis approach, applying time-dependent COX frailty models, in which the target variable is the time to dropout of students within the first three years after the enrolment. Student careers are tracked over time, collecting time-dependent information. Results show that first year information is already powerfully predictive of the time to dropout and that dropout trends differ across degree courses and student profiles.