Feature Selection for Self-Organizing Map

In this paper, we present a new heuristic measure for optimizing database used as input layer of Self Organizing Map (SOM). This heuristic called Hl-SOM (Heuristic Input for SOM) consists of selection of variables for clustering with SOM algorithm. HI-SOM allows to identify and to select important variables in the feature spaces. Thus, we eliminate redundant variables and those do not contain enough relevant information. The proposed measure is used in SOM learning algorithm in order to reduce the database dimension. Hence, HI-SOM select the important variables to train the "best" SOM. We illustrate this method with three databases from public data set repository. We show the effectiveness to identify the important variables which gives homogenous clusters.