A Weighted Object-Cluster Association-Based Ensemble Method for Clustering Undergraduate Students

Abstract

Clustering ensemble methods have received much attention for better clustering quality and robustness as they can exploit the knowledge discovered in their base clusterings. In order to obtain better clusterings on educational data, in this work, we propose a novel clustering ensemble method as the first ensemble-based solution to an educational data clustering task. Different from the existing ensemble methods, our method is based on a weighted object-cluster association matrix. We define this association matrix as a synthesis of the base clusterings. It can capture not only the inherent structure of the data via base clusterings but also the discrimination between the objects via their cluster representatives. As a result, our method effectively groups the students into the clusters each of which have the most similar students based on study performance. This is confirmed by better Normalized Mutual Information values from the experiments on the real educational data sets and the popular Iris data set.