A comparison of the utility of data mining algorithms in an open distance learning context

• Published in: South African journal of higher education Vol. 32; no. 4; pp. 81 - 95
• Main Authors: Fynn, A, Adamiak, J
• Format: Journal Article
• Language: English
• Published: Higher Education South Africa (HESA) 01-09-2018
• Subjects: Educational Data Mining; J48; Learning analytics; Logistic regression; Student success; WEKA
• ISSN: 1011-3487; 1753-5913; 1753-5913
• DOI: 10.20853/32-4-2473

The use of data mining within the higher education context has, increasingly, been gaining traction. A parallel examination of the accuracy, robustness and utility of the algorithms applied to data mining is argued as a necessary step toward entrenching the use of EDM. This article provides a comparative analysis of various classification algorithms within an Open Distance Learning institution in South Africa. The study compares the performance of the ZeroR, OneR, Naïve Bayes, IBk, Simple Logistic Regression and the J48 in classifying students within a cohort over an eight-year time span. The initial results appear to show that, given the data quality and structure of the institution under study, the J48 most consistently performed with the highest levels of accuracy.