Enhancement Clustering Evaluation Result of Davies-Bouldin Index with Determining Initial Centroid of K-Means Algorithm

Abstract

Abstract K-Means is one of the most popular clustering algorithms because it is easy and simple when implemented. However, clustering results from K-Means are very sensitive to the selection of initial centroid. Better clustering results are often obtained after several experiments. In this study, Sum of Squared Error (SSE) was used as an approach to determine initial centroid of K-Means algorithm. If the SSE value is smaller then the data in one cluster will be more homogeneous and certainly give a good cluster result. In this study, Sum of Squared Error (SSE) was used as an approach to determine initial centroid of K-Means algorithm. Testing was performed on 3 datasets and the number of clusters 2, 3 and 4. From the test, the average value of Davies-Bouldin Index (DBI) for 3 datasets was 0.2427, while the Simple determine initial centroid of K-Means algorithm obtained an average DBI value of 0.2805. These results prove that clustering with method of determining initial centroid of K-Means algorithm based on Sum of Squared Error minimum able to improve clustering result and enhance DBI value obtained by simple determine initial centroid of K-Means algorithm.