Improved student dropout prediction in Thai University using ensemble of mixed-type data clusterings

• Published in: International journal of machine learning and cybernetics Vol. 8; no. 2; pp. 497 - 510
• Main Authors: Iam-On, Natthakan, Boongoen, Tossapon
• Format: Journal Article
• Language: English
• Published: Berlin/Heidelberg Springer Berlin Heidelberg 01.04.2017; Springer Nature B.V
• Subjects: Academic achievement; Algorithms; Artificial Intelligence; Classification; Clustering; Colleges & universities; Complex Systems; Computational Intelligence; Control; Data collection; Data mining; Decision trees; Engineering; Feature selection; Learning; Mechatronics; Original Article; Pattern Recognition; Principal components analysis; Robotics; Student retention; Students; Systems Biology; Variables; Student dropout; Educational data mining; Feature transformation; Classification; Ensemble clustering
• DOI: 10.1007/s13042-015-0341-x
• ISSN: 1868-8071

Increasing student retention has been a common goal of many academic institutions, especially in the university level. The negative effects of student attrition are evident to students, parents, university and the society as a whole. The first-year students are at the greatest risk of dropping out or not completing their degree on time. With this insight, a number of data mining methods have been developed for early detection of students at risk of dropout, hence the immediate application of assistive measure. As compared to western countries, this subject has attracted only a few studies in Thai university, with educational data mining being limited to the use of conventional classification models. This paper presents the most recent investigation of student dropout at Mae Fah Luang University, Thailand, and the novel reuse of link-based cluster ensemble as a data transformation framework for more accurate prediction. The empirical study on mixed-type data collection related to students’ demographic detail, academic performance and enrollment record, suggests that the proposed approach is usually more effective than several benchmark transformation techniques, across different classifiers.