Applicability of K-medoids and K-means algorithms for segmenting students based on their scholastic performance

Abstract

In this paper literature was surveyed to find popular clustering techniques used by researchers in recent times to predict academic performance. We obtained a trend that the K-means algorithm is particularly popular among researchers because of its simplicity and scalability, and in other studies K-medoids algorithm was selected as it is less affected by outliers. On the basis of these observations these two clustering algorithms were implemented in Python, on student dataset of undergraduate students from a higher education institute. Two different clusters were obtained which segment students based on their academic performances in the previous two exams. The clusters obtained by have high accuracy score and K-medoids cluster centroids have taken exact values of marks obtained by students whereas K-means centroid value is a round off. The K-means clustering is also affected by the presence of outliers in the student dataset.

Subject Classification: (2010):

• 68P99

Keywords:

• Educational data mining
• Clustering
• K-means
• K-medoids
• Academic performance
• Outliers