Resumen de A weighted voting summarization of SOM ensembles

Bruno Baruque Zanón, Francisco Martínez-Álvarez

• Abstract: Weighted Voting Superposition is a novel summarization algorithm for the results of an ensemble of Self-Organizing Maps. Its principal aim is to achieve the lowest topographic error in the map in order to obtain the best possible visualization of the internal structure of the data sets under study. This is done by means of a weighted voting process between the neurons of the ensemble maps in order to determine the characteristics of the neurons in the resulting map. The algorithm is applied in this case to the most widely known topology preserving mapping architecture: the Self- Organizing Map. A comparison is made between the novel fusion algorithm presented in this work and other previously devised fusion algorithms, along with a new variation of those algorithms, called Ordered Similarity. Although a practical example of the new algorithm was introduced in an earlier work, a rigorous description and analysis is presented here for the first time by comparing the performance of the aforementioned algorithms in relation to three well-known data sets (Iris, Wisconsin Breast Cancer and Wine) obtained from Internet repositories. The results show how this novel fusion algorithm outperforms the other fusion algorithms, yielding better visualization results for ensemble summarization of maps.