CMStatistics 2016: Start Registration
View Submission - CMStatistics
Title: Comparing robust fuzzy methods for clustering non-precise data Authors:  Ana Belen Ramos-Guajardo - University of Oviedo (Spain) [presenting]
Paolo Giordani - Sapienza University of Rome (Italy)
Abstract: In many practical situations the data are not precise. The imprecision of the data can be managed by means of fuzzy sets. This type of data is characterized by a complex structure and, for this reason, there exist different kinds of contamination in this context. There are several proposals of robust methods for clustering fuzzy data. A type of `robustification' is the use of medoids. Another approach consists in trimming the data. In detail, the outliers are trimmed and not used in the clustering procedure. A further proposal is to add a noise cluster, that is not a proper cluster, containing all the contaminated data. Finally, an alternative approach is the possibilistic one. In this case the membership degree is only based on the distance between the observation and the centroid. Hence, an outlier is characterized by having low membership degrees to all the clusters. We compare all the above mentioned methods by means of simulation and real-case studies in order to analyze their drawbacks and benefits.