Comparative Performance Analysis of Clustering Techniques in Educational Data Mining

Clustering analysis provides a useful way to group objects without having previous knowledge about the data being analysed. In this paper, we first survey the research done on clustering analysis in education and identify the algorithms used. We then present a case-based experiment to show the relative performance of clustering algorithms with Learning Management System log data. We compare partition-based (K-Means), density-based (DBSCAN) and hierarchical (BIRCH) methods to determine which technique is the most appropriate for performing clustering analysis within the LMS. We conclude by showing that partition-based methods produce the highest Silhouette Coefficient values and the better distribution amongst the clusters. The BIRCH algorithm also performs fairly well and can act as a good starting point to find cluster groups in new datasets as the algorithm does not required that the number of clusters be specified a priori.