Online Robust Fuzzy Clustering of Data with Omissions Using Similarity Measure of Special Type

Abstract

The task of clustering is important and the most difficult part of the overall problem of Data Mining because it is based on the self-learning paradigm, i.e. implies the absence of pre-tagged training sample. In real conditions, this task is complicated by the fact that in having data arrays some of the observations can be corrupted by anomalous outliers and some - contain missing data, that is, the “object-property” table has “empty” cells. In addition, data can be arrive in online mode on processing, especially for tasks related with Data Stream Mining and Big Data. In the paper the problem of fuzzy adaptive online clustering of data distorted by outliers that are sequentially fed to the processing when the original sample volume and the number of distorted observations are apriori unknown is considered. The probabilistic and possibilistic adaptive online clustering algorithms for such data, that are based on the similarity measure of a special type that weaken or overwhelming outliers are proposed. The computational experiment confirms the effectiveness of approach under consideration.