Contribuição ao estudo de fusão de mapas auto organizáveis de Kohonen com ponderação por meio de índices de validação de agrupamentos

The amount of collected and stored information is growing every day in several areas of knowledge and data mining techniques are applied to these datasets in order to extract useful knowledge. One or another algorithm, or the same algorithm with different attributes, can lead to different results due to the dataset diversity. To solve this problem, machines committees methods were developed. A machine committee is a set of neural networks working independently and the results are combined into a single output, achieving a better generalization. The purpose of this work is to develop a new method for Kohonen maps ensemble, where the maps fusion is weighted by cluster validation indices and is suitable for equal size maps fusion and for different size maps fusion. The proposed algorithm has been tested in multiple data sets from the UCI Machine Learning Repository and Fundamental Clustering Problems Suite. Computer simulations show the proposed method is able to reach encouraging results, obtaining raising performance compared with a single Kohonen map.