Applying Fuzzy Possibilistic Methods on Critical Objects

Providing a flexible environment to process data objects is a desirable goal of machine learning algorithms. In fuzzy and possibilistic methods, the relevance of data objects is
evaluated and a membership degree is assigned. However, some critical objects objects have the potential ability to affect the performance of the clustering algorithms if they remain in a specific cluster or they are moved into another. In this paper we analyze and compare how critical objects affect the behaviour of fuzzy possibilistic methods in several data sets. The comparison is based on the accuracy and ability of learning methods to provide a proper searching space for data objects. The membership functions used by each method when dealing with critical objects is also evaluated. Our results show that relaxing the conditions of participation for data objects in as many partitions as they can, is beneficial.