A Theoretical Model to Predict Undergraduate Learner Attrition using Background, Individual, and Schooling Attributes
Researcher: Noluthando Mngadi, University of the Witwatersrand, Johannesburg
Supervisor: Dr Ritesh Ajoodha and Dr Ashwini Jadhav, University of the Witwatersrand, Johannesburg
There is a growing concern around student attrition worldwide, including South African universities. More often than not, the reasons for students not completing their degree in the allocated time frame include academic reasons, socio-psycho factors, and lack of effective transition from the secondary education system to the tertiary education systems. To overcome these challenges, the tertiary educational institutions endeavor to implement interventions geared toward academic success. One of the challenges, however, is identifying the vulnerable students in a timely manner. This study therefore aims to predict student performance by using a learner attrition model so that the vulnerable students are identified early on in the academic year and are provided support through effective interventions, thereby impacting student success positively. Predictive machine learning methods, such as support vector machines, decision trees, and logistic regression, were trained to deduce the students into four risk-profiles. A random forest outperformed other classifiers in predicting at-risk student profiles with an accuracy of 85%, kappa statistic of 0.7, and an AUC of 0.95. This research argues for a more complex view of predicting vulnerable learners by including the student’s background, individual, and schooling attributes.
Keywords: Attrition, At-risk, Machine learning