Clustering Student Learning Activity Data

We show a variety of ways to cluster student activity datasets using different clustering and subspace clustering algorithms. Our results suggest that each algorithm has its own strength and weakness, and can be used to find clusters of different properties. 1 Background Introduction Many education datasets are by nature high dimensional. Finding coherent and compact clusters becomes difficult for this type of high dimensional data. Subspace clustering was proposed as a solution to this problem [1]. Subspace clustering searches for compact clusters embedded within subsets of features, and it has proven its effectiveness in domains that have high dimensional datasets similar to educational data. In this paper, we will show that different clustering and subspace clustering algorithms produce clusters of different properties, and all these clusters help the instructor assess their course outcomes from various perspectives. 2 Clustering Student Activity Data We assume that datasets are in the following format: each row represents one student record, and each column measures one activity that students participate in. Our test data contains 30 students with 16 activities, and 7 students failed this class. The final grade is the weighted average from the scores in all 16 activities. 2.1 Student clusters Student clusters consist of groups of students who demonstrate similar learning curves throughout the whole course. These clusters are helpful to identify key activities that differentiate successful students from those who fail the course. We applied the SimpleKMeans from Weka [2] to the test dataset with k being set to 2. The results show that cluster1 contains 6 out of 7 students who failed the course, and cluster2 contains 24 students among whom 23 passed the course. One student who failed was clustered into cluster2, and we found out that this student’s composite final score is 58%, which lies right on the boundary of passing/failing threshold. 2.2 Activity clusters Here we focus on finding groups of activities in which all students demonstrate similar performance. For example, we may find a group of activities where all students show