Reduksi Dimensi untuk Meningkatkan Kinerja Pengklasteran Perilaku Siswa pada Sistem e-Learning

The corona pandemic has changed the learning process from face-to-face (offline) to online learning. However, this online learning has caused difficulties in monitoring student behavior by teachers due to reduced direct interaction. Additionally, students often feel isolated. Therefore, this situation causes failure in their learning achievement. This problem encourages a lot of research on modeling related to student behavior. However, previous research did not focus much on improving the model's performance or system being built. In fact, the performance of this model significantly affects the result’s quality of this student behavior mapping. Therefore, this study focuses on improving the performance of student behavior clustering when they interact with the e-Learning system. Performance improvement was made by reducing dimensions of student data with Principal Component Analysis (PCA). Furthermore, two techniques for the centroid initialization were explored to obtain optimal results: random and K-means++. For measuring cluster quality, this study employed the silhouette index. The experimental results show that the clusters with the highest quality are achieved by applying PCA with seven components. In addition, the cluster number for all centroid initialization techniques is three to four. This quality cluster can assist teachers in monitoring student behavior in the e-Learning system.