Comparison of Clustering Algorithms for Learning Analytics with Educational Datasets

Abstract

Learning Analytics is becoming a key tool for the analysis and improvement of digital education processes, and its potential benefit grows with the size of the student cohorts generating data. In the context of Open Education, the potentially massive student cohorts and the global audience represent a great opportunity for significant analyses and breakthroughs in the field of learning analytics. However, these potentially huge datasets require proper analysis techniques, and different algorithms, tools and approaches may perform better in this specific context. In this work, we compare different clustering algorithms using an educational dataset. We start by identifying the most relevant algorithms in Learning Analytics and benchmark them to determine, according to internal validation and stability measurements, which algorithms perform better. We analyzed seven algorithms, and determined that K-means and PAM were the best performers among partition algorithms, and DIANA was the best performer among hierarchical algorithms