Improving the Clustering Performance of the K-Means Algorithm for Non-linear Clusters

Abstract

K-means clustering is known to be the most traditional approach in machine learning. It's been put to a lot of different uses. However, it has difficulty with initialization and performs poorly for non-linear clusters. Several approaches have been offered in the literature to circumvent these restrictions. Kernel K-means (KK-M) is a type of K-means that falls under this group. In this paper, a two-stepped approach is developed to increase the clustering performance of the K-means algorithm. A transformation procedure is applied in the first step where the low-dimensional input space is transferred to a high-dimensional feature space. To this end, the hidden layer of a Radial basis function (RBF) network is used. The typical K-means method is used in the second part of our approach. We offer experimental results comparing the KK-M on simulated data sets to assess the correctness of the suggested approach. The results of the experiments show the efficiency of the proposed method. The clustering accuracy attained is higher than that of the KK-M algorithm. We also applied the proposed clustering algorithm on image segmentation application. A series of segmentation results were given accordingly.