Feature Transformation Over RBF Network For Improving The Performance Of K-Means Clustering Algorithm
The K-means clustering is a popular clustering algorithm in the pattern recognition and machine learning
communities. It has been used in a vast of applications. However, it suffers from the problem of initialization and poor
performance for the non-linear clusters. To overcome these limitations, several attempts have been proposed in literature.
Kernel K-means can be considered in that category. In this paper, we proposed a two-step scheme to improve the
performance of the K-means algorithm. In the first step, the transformation of the low dimensional input space to a high
dimensional feature space is carried out. This transformation is carried out via the hidden layer Radial basis function (RBF)
network. The second step of our proposal is standard K-means algorithm. To analyze the validity of the proposed scheme,
we present experimental results that compare the kernel K-means on artificial data sets. The experimental results show that
this simple new scheme is efficient in clustering non-linearly separable clusters. The obtained accuracy is higher than the
kernel K-means algorithm.
Index Terms- K-Means Clustering, Feature Transformation, RBF Network.