K-means-based fuzzy classifier design

In this paper, a method based on the K-means algorithm is proposed to efficiently design a fuzzy classifier so that the training patterns can be correctly classified by the proposed approach. In this method, the K-means algorithm is first used to partition the training data for each class into several clusters, and the cluster center and the radius for each cluster are calculated. Then, a fuzzy system design method that uses a fuzzy rule to represent a cluster is proposed such that a fuzzy classifier can be efficiently constructed to correctly classify the training data. The proposed method has the following features: 1) it does not need prior parameter definition; 2) it only needs a short training time; and 3) it is simple. Finally, two examples are used to illustrate and examine the proposed method for the fuzzy classifier design