Abstract
Student retention is an increasingly important yet complex issue facing universities. Improving retention performance is part of a multidimensional and deeply nested system of relationships with multiple hypothesised drivers of attrition at various sample sizes, population clusters and timescales. This paper reports on the use of a self organising data technique, Kohonen’s Self Organising Map, to explore the potential retention drivers in a large undergraduate student population in Western Australia over a six-year period. The study applied the self-organizing method to two point-in-time data sets separated by 18 months and was able to identify a number of distinct attrition behaviour profiles appropriate for creating new tailored intervention.
Keywords
Attrition, retention, predictive models, machine learning, educational data mining, learning analytics.