A theoretical model to predict undergraduate attrition based on background and enrollment characteristics

Date
2020
Authors
Mathye, Macdaline Raisibe
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Developing graduate readiness amongst students who enters university with risk factors is one of the greatest challenges of institutions. Evidence that students with risk profiles are not likely to seek assistance when required complicates the problem. In this work we aim to identify the profiles of students with attributes indicating learner vulnerability .A synthetic higher education dataset from 2008-2018 was used for the purpose of this research. We follow the conceptual framework by Tinto (1975) to deduce student attrition. The features considered were academic courses, grade 12 marks, back-ground information, individual attributes and respective outcomes for science students. To identify profiles of vulnerable students, several ma-chine learning classification models to deduce the learner into four risk classes: Lowest Risk, Medium risk, High risk and Highest risk were used. The analysis used various predictive models: Random Forests, Decision trees, Support vector Machines, Bayesian Network classifier and multinomial Logistics regression. Effectiveness of each model was tested through 10-Fold Cross Validation and all the hyperparameters were tuned. The Random Forest performed the best with an accuracy of 73% and the least predictive model with 63% was the Multinomial Logistic Regression. The major contribution of this report are: a) a comparison of predictive models to calculate the probability of a learner’s risk profile, by contextualizing the students synthetic background, individual and schooling data. b) a ranking of employed features according to their entropy to correctly classify the class variable
Description
A research report submitted in partial fulfillment of the requirements for the degree of Master of Science in the field of e-Science in the School of Computer Science and Applied Mathematics, University of the Witwatersrand, 2020
Keywords
Citation
Collections