Determination of Initial K-means Centroid in the Process of Clustering Data Evaluation of Teaching Lecturers

摘要

Decision making about microteaching for lecturers in ITTP with the low teaching quality is only based on three lowest order from teaching values. Consequently, the decision is imprecise, because there is possibility that the lecturers are not three. To get the precise quantity, an analysis is needed to classify the lecturers based on their teaching values. Clustering is one of analyses that can be solution where the popular clustering algorithm is k-means. In the first step, the initial centroids are needed for k-means where they are often randomly determined. To get them, this paper would utilize some preprocessing, namely Silhouette Density Canopy (SDC), Density Canopy (DC), Silhouette (S), Elbow (E), and Bayesian Information Criterion  (BIC). Then, the clustering results by using those preprocessing were compared to obtain the optimal clustering. The comparison showed that the optimal clustering had been given by k-means using Elbow where obtain four clusters and 0.6772 Silhouette index value in dataset used. The other results showed that k-means using Elbow was better than k-means without preprocessing where the odds were 0.75. Interpretation of the optimal clustering is that there are three lecturers with the lower teaching values, namely N16, N25, and N84.