Extracting Rules from Optimal Clusters of Self-Organizing Maps

Self-organizing map (SOM) neural networks have been successfully applied to solve classification and clustering problems. However, while most SOM models pursue their results as accurately as possible, they ignore the importance of understanding and explanation. This paper first finds the optimal solution for the number of SOM clusters by using the technique of particle swarm optimization (PSO) and then generates clustering rules by extracting implicit knowledge from a one-dimensional SOM neural architecture. The experimental results show that rules extracted by our method produce an improvement in performance compared with other rule extraction models. Our proposed approach is able to equip the self-organizing map with an explanatory capability through the use of rules.