An improved multi-SOM algorithm

Clustering methods aim to obtain homogeneous partitions of objects while promoting the heterogeneity between these partitions. Every clustering approach such as hierarchical, partitioning and neuronal methods has eventually its advantages and limits. We focus on neuronal methods as they overcome the limits of the hierarchical and partitioning methods and they are the most appropriate clustering approaches to use for a large number of data. In this work, we propose a multi-SOM algorithm using a different evaluation criterion. Thus, a review of the evaluation measures proposed in the literature is necessary. Nevertheless, multi-SOM method along with its strength and efficiency in the delimitation of clusters has also a limit at the stop condition.