An improved rough clustering using discernibility based initial seed computation

Abstract

In this paper, we present the discernibility approach for an initial seed computation of Rough K-Means (RKM). We propose the use of the discernibility initial seed computation (ISC) for RKM. Our proposed algorithm aims to improve the performance and to avoid the problem of an empty cluster which affects the numerical stability since there are data constellations where | Ck| = 0 in RKM algorithm. For verification, our proposed algorithm was tested using 8 UCI datasets and validated using the David Bouldin Index. The experimental results showed that the proposed algorithm of the discernibility initial seed computation of RKM was appropriate to avoid the empty cluster and capable of improving the performance of RKM. © 2010 Springer-Verlag.