Improvement of FCM neural network classifier using K-Medoids clustering

Abstract

Floating Centroids Method (FCM) is a new method to improve the performance of neural network classifier. But the K-Means clustering algorithm used in FCM is sensitive to outliers. So this weakness will influence the performance of classifier to a certain extent. In this paper, K-Medoids clustering algorithm which can diminish the sensitivity to the outliers is used to partition the mapping points into some disjoint subsets to improve FCM's robustness and performance. Some data sets from UCI Machine Learning Repository are employed in our experiments. The results show a better performance for the FCM using our improved method.