Predicting Student Dropout Using Demographic Indicators: Internal Report and Interactive Dashboard

Background: The rate of dropout from degree programs is a significant issue for university administrators and academics both globally and at La Trobe. If at-risk students could be identified before commencement of their first year, then resources could be prioritised towards students likely to require the most support. As such, this investigation sought to i) model dropout from several degrees in the SAHHSS using only student entry demographic data, and ii) demonstrate a common and intuitive machine learning procedure for classifying student dropout.

Methods: Descriptive statistics for each demographic indicator were acquired for the 2019 cohort for the following degrees: HBSES, HBSCD, HBHN, HZSK, HMSPH. A chi-squared automatic interaction detection (CHAID) algorithm was employed to model relationships between the demographic indicators and dropout where sufficient cases were present.

Findings: Sufficient dropouts and enrolments to perform the analysis were present for HBSES, HBHN, and HZSK. HBSCD had only nine enrolments, while HMSPH demonstrated no dropouts. Of the eight input variables, the CHAID was successfully able to classify dropout using only three to four variables, with Region and ATAR featuring prominently. The models achieved an 87-89% classification accuracy.

Conclusions and Applications: These findings enable academics and administrators of the examined degrees to target early stage student retention interventions towards those most at risk. Further, because the CHAID successfully identified multiple patterns of indicators to classify dropout using only three or four key variables, academics can therefore consider this analysis procedure an effective method to understand the dropout characteristics of a cohort.