EFFICIENT CLUSTERING OF MULTIDIMENSIONAL DATA.

An efficient algorithm for the clustering of multidimensional data is developed. The proposed technique is based on the possible improvement of the solution when a local minimum solution is obtained. A search for an improving point is proposed by considering the extreme points of the problem constraints which are adjacent to the current solution point produced by the standard version of the K-MEANS algorithm. The proposed algorithm proved to be effective on two accounts; the computation time and the resulting value of the error sum of squares. Experimental results with comparisons to illustrate this fact are reported.