A robust fuzzy clustering method with outliers influence free
Li-Jen Kao, Yo‐Ping Huang · 2012
Fuzzy C-means algorithm (FCM) is a method of clustering which allows a point data to belong to two or more clusters. FCM algorithm suffers from outliers or noise because of the sum of membership values for an outlier point in all the clusters still being one. In this paper, an adapted FCM algorithm is proposed not only to detect the outliers but also remove the outliers to make FCM method robust. The algorithm gets a point's outlier degree on a certain cluster according to its Euclidean distance to that cluster and if the outlier degree is greater than a pre-defined threshold, that point will be assigned 0 membership value in that cluster. This makes the outliers influence free on cluster centers calculation. A point is a true outlier if all of its cluster's outlier degrees are greater than a pre-defined threshold. The experiments show that the proposed algorithm can get new cluster centers in a more efficient way.