CSOM: Self-organizing map for continuous data

Nowadays, lots of data is being collected for different industrial and commercial purposes, where the aim is to discover useful patterns from data which leads to discovery of valuable domain knowledge. Unsupervised learning is a useful method for these tasks as it requires no target class and it clusters the feature values that occur frequently together. Clustering methods have been successfully used for this task due to the powerful property of creating spatial representations of the features and the abstractions detected from the input space. Self-Organizing Map (SOM) [6] is one of the most popular clustering techniques where abstractions are formed by mapping high dimensional input patterns into a lower dimensional set of output clusters. Most of the current uses of SOM for this task concentrated on clustering categorical features [2, 4]. In this paper we present a new learning mechanism for Self-Organizing Map which is useful when the aim is to extract patterns from a data set characterized by continuous input features. Furthermore the method used for network pruning and rule optimization is described.