Clustering students based on grammatical errors for on-line education

Abstract

Learning Management System (LMS) is an educational solution created for people who need flexibility regarding time and place. The problem of this kind of tool primarily concerns the difficulty in identifying which students have learned the content correctly. This paper aims to analyze the performance of a group of distance learning students regarding grammar errors in two different terms of an undergraduate course. Our hypothesis relies on the existence of different characteristics that emerge from subgroups of students with similar difficulties. This division can help tutors in educational platforms to develop specific recommendations tasks for each group of students. A previous work applied the well-known K-means algorithm to cluster the groups, but in that paper, we fixed the number of clusters. Therefore, we carried out a methodology to find the best number of clusters to be used in K-means for this problem. Moreover, we also applied the Fuzzy C-means to tackle the clustering problem and analyzed the results obtained by both algorithms using the well-known metrics in the literature (Gap Statistic and Davies-Bouldin) to assess the quality of the obtained groups. The experimental results showed that Fuzzy C-means approach outperforms the K-means algorithm. Moreover, the application of the Spearman Correlation on each group expose several differences, relations and similarities between groups and inside each one.