Novel method to optimize the architecture of Kohonen´s topological maps and clustering

Clustering methods are one of the most important tools used in different areas by researchers. The self-organizing map network is one of the most popular neural networks which was designed for solving problems that involve tasks such as clustering, visualization, and abstraction. Specially, It provides a new strategy of clustering using a competition and co-operation principal. However, the optimal number of neurons and their initial weight vectors in the map is still a big problem in the literature. These parameters have a great impact on the learning process of the network. In this paper we extend the original Kohonen network. First, the task consists of generating a heuristic method before the learning phase of the network. The main goal of this method is looking for the initial parameters of the map. That is, finding centroids of the most homogenous areas of data set. Second, the learning phase of Kohonen is run and the evaluation of the obtained clustering by the map. The two phases are repeated until a number of iteration is satisfied. We compare the result of the proposed method with that of the original Kohonen network. We further experiment the applicability and the performance of the method on various data sets of various dimensions known in the literature. The result shows that the proposed method is able to produce better clustering results than the traditional topological map.