Self-Organizing Maps for High Dimensional Feature Space Problems

Self Organizing Map SOM is one of the wide spread neural network approaches which enable feature extraction visualization in pattern recognition and data mining problems. However two major problems have been observed when SOM is used in high dimensional feature space problems: the recognition rate is dropped when unseen data are used and expensive computational cost when using large data sets. In this book we develop new hierarchical search algorithm for SOM. Several experiments are conducted for some pattern recognition applications showed the superiority of the proposed SOM over traditional SOM and other related approaches.