mass-grsom: a self organizing map for rule extraction using a novel parameterized metric distance

The objective of this work is the extension of a specific family of Self Organizing Maps named granular Self-Organizing Maps or grSOM for short, by using a novel parameterized metric distance based on lattice theory. The resulting Self-Organizing map, namely, mass-grSOM is applicable beyond Rn to Fn where Fn denotes the set of fuzzy interval numbers (FINs). The proposed Self-Organizing Map, used here for linguistic modeling applications, describes efficiently the input space of a system, by calculating a set of multi dimensional FINs from a number of input/output observation data. The mass-grSOM, is created in two separate stages. Firstly, an optimal set of multidimensional FINs, which represent local probability distributions, is calculated using Kohonen’s self organization principles, through grSOM. Secondly, a genetic algorithm is employed to adjust the parameters of a novel metric distance, improving the classification performance and leaving the initial location of FINs, intact. Experimental results on a real world classification benchmark shown that mass-grSOM outperforms alternative classification schemes, refereed to the literature.