Student modeling using principal component analysis of SOM clusters

Abstract

Adaptive hypermedia learning systems aim to improve the usability of hypermedia by personalizing domain knowledge to the students' needs (represented by the student model). This study investigates student modeling via machine-learning techniques. Two techniques are applied and compared to provide meaningful analysis and class labels of the student clusters. The first technique is clustering of the student data set using principal component analysis. The second technique involves two-levels of clustering: the self organizing map at the first level and principal component analysis at the second level. Cluster analysis via these two techniques determine the number of clusters, the class labels based on the degree of variance and eigenvectors, which can represent the knowledge states of each cluster or group of students. It is found that implementing the self-organizing map as a preprocessor to principal component analysis improves the quality of cluster analysis. Findings are supported by experimental results.