Performance Improvement of Grid Mapping K-Means with the Average Value at Grid Point

Abstract

K-Means is a method that is well-known for its ability to handle large datasets, but is often stuck in a local optima state. This issue happens because K-Means generally uses random numbers to serve as the center point (centroid) of each cluster, and places each instance based on the proximity of the distance using Euclidean Distance. Hence, the concept of density parameter was developed, which tries to determine the ideal centroid based on the determination of several grid points in each existing cluster. The concept is known as Grid Mapping. In simple terms, Grid Mapping K-Means is a method in which each cluster is divided into several grid points, and then the centroid is randomly determined from each existing grid point. Further, the grid point with the largest number of instance is used as the initial centroid of each cluster. However, this method certainly has a weakness, since there is a possibility that the initial centroid on the generated grid point is not the best initial centroid. Therefore, this study was conducted to test the determination of initial centroid by Grid Mapping K-Means.

Analisis Algoritma K-Means dan Davies Bouldin Index dalam Mencari Cluster Terbaik Kasus Perceraian di Kabupaten Kuningan

“…K-Means Clustering dapat diterapkan pada kasus dengan jumlah objek yang sangat banyak [12]. Algoritma K-means adalah algoritma pengelompokan iteratif sederhana [13] dan yang paling populer digunakan dengan menggunakan jarak sebagai metriknya [14].…”