Comparison of Algorithms K-Means and DBSCAN for Clustering Student Cognitive Learning Outcomes in Physics Subject

Abstract

Clustering is an activity of grouping data into the same group based on similarity. The purpose of the study is to cluster and determine student cognitive learning outcomes characteristics. Cluster analysis was conducted on student cognitive learning outcomes using algorithms K-Means and DBSCAN. Both algorithms are appropriate to have been applied to the overlapping data such as student learning outcomes data. Also, their advantages are scaling large datasets and outliers. The data used in this study is student cognitive learning outcomes - final and mid-term exams grade X in physics subject. Applying the two proposed algorithms K-Means and DBSCAN, the best cluster algorithm to have been used for clustering analysis is K-Means which is based on the highest silhouette score of 0.43, while the silhouette score of DBSCAN is 0.39 respectively. Using the best cluster, the K-Means algorithm, found two types of clusters – cluster 1 consists of 132 students who have a high average score, and cluster 2 shows 183 students who have a low average score in both final and mid-term exams respectively. From the analysis results, most students still have low cognitive learning outcomes in physics subject.