Model-based clustering with mclust R package: multivariate assessment of mathematics performance of students in Qatar

This study demonstrates how model-based clustering can be undertaken using mclust, a contributed R package, to examine factors influencing mathematics performance of high school students in Qatar. Although there are numerous cluster analysis approaches, this paper highlights the intricacies, assumptions, limitations, benefits and pitfalls of clustering using a model-based approach, and how the inherent inadequacies of other clustering approaches can be better explored using model-based methods. Moreover, this paper demonstrates how the mclust package can be used to concurrently analyse and compare different models, in order to select the preferred clustering model according to the Bayesian information criterion, and to estimate parameters of the associated model using maximum likelihood estimation. The benefit of selecting a prior to avoid model-based clustering estimation singularity- and degeneracy-related issues offers an alternative approach to improve the rate of convergence. The results from applying model-based clustering using mclust to educational data that examines the mathematics performance of secondary students in Qatar will be used to identify factors that influence mathematics performance for different clusters of students, to help facilitate potential adoptions of the most appropriate remedial teaching strategies to implement to enhance learning. Furthermore, the results can help teachers to identify groups of students whose performance in different subject areas is likely to be affected by certain factors, thereby helping them to reduce potentially undesirable learning outcomes.