A behavioral study of learning in standard SOM and in S-SOM

In this paper we establish the completion of a previously published work, in part by the same authors, in which we proposed a novel learning algorithm involving self organizing map´s (SOM) internal structure in the learning process. We present a statistical and a behavioral study of our proposed solution, and confirm its results on a breast cancer classification application. After establishing the mathematical foundations of the passage of input vectors, we show how to conclude better initial conditions leading to the generation of a unique feature map in standard SOM.