IMPROVING THE PERFORMANCE OF K-MEANS ALGORITHM USING AN AUTOMATIC CHOICE OF SUITABLE CODE VECTORS AND OPTIMAL NUMBER OF CLUSTERS
Mohamed Ettaouil, Es-safi Abdelatif, Ibn Abdelah · 2013
The automatic clustering is a useful tool for data-mining. It‘s a daily necessity for the searcher whatever his specialty. Indeed because of the huge amount of information available on the web-site, the access to relevant information in a suitable time is a difficult task. By grouping those informations in clusters this problem can be surmounted. Many clustering methods exist in the literature but the efficient ones suffer from some drawbacks. The main of them follows from the initialization phase which is performed randomly. Among these algorithms we find the k-means(deterministic and probabilistic version) and the clustering method based on Gaussian mixture. In these algorithms the initial parameters including the number of cluster are chosen randomly. Consequently an improper choice leads to poor clusters. In this paper we propose an approach attempting to overcome these problems. In this method the initial parameters are automatically and suitably identified. To this end, the structure of data is investigated in each iteration. To validate the proposed method a number of experiments are performed.